{"dataset":{"name":"Speech recognition word error rates","title":"Speech Recognition Accuracy: Word Error Rates","question":"How accurate is speech recognition, by system, test set, language and speaking condition?","url":"https://yapapp.cc/data/asr-accuracy","measure":{"label":"Word error rate","unit":"percent","short":"WER %","decimals":2,"lowerIsBetter":true},"dimensions":[{"key":"system","label":"System"},{"key":"kind","label":"Kind","values":["Open model","Cloud service","Built-in dictation","Research system","Human transcribers"]},{"key":"benchmark","label":"Test set"},{"key":"language","label":"Language"},{"key":"condition","label":"Speech","values":["Clean read speech","Conversational","Noisy or distant","Accented or dialect","Children","Atypical or disordered speech","Specialist vocabulary","Mixed benchmark"]},{"key":"year","label":"Year"}],"include":"Word error rates reported for a named system on a named test set or recording collection in a stated language, from papers, public leaderboards, standards bodies, and independent evaluations that publish their method. Human transcriber error rates on the same tests are included for comparison.","exclude":"Accuracy percentages a vendor states without a test set, relative improvements without the absolute rate, and character error rates (a different measure).","method":"Rows come from the facts already cited across the library, from a structured search of speech recognition research (OpenAlex, Semantic Scholar and arXiv), and from public leaderboards. Each figure is stored as a fact with the exact words of its source, checked automatically against the live page. Rates stated as fractions (0.12) are converted to percent (12).","limitations":"Word error rates depend on the test set, how text is normalized before scoring, and whether a language model was used, so compare systems only on the same test set. Lower is better; a rate can exceed 100 percent.","sources":32,"updated":"2026-10-19","license":"https://creativecommons.org/licenses/by/4.0/","credit":"Yap (2026). Speech recognition word error rates [Data set]. https://yapapp.cc/data/asr-accuracy"},"rows":[{"row":"Canary 1B Flash, LS test-clean English","value":1.48,"low":null,"high":null,"as_written":"1.48","unit":"percent","system":"NVIDIA Canary 1B Flash","kind":"Open model","benchmark":"LibriSpeech test-clean","language":"English","condition":"Clean read speech","year":null,"conversion":null,"claim":"NVIDIA Canary 1B Flash had a word error rate of 1.48 percent on the LibriSpeech test-clean test set in the Open ASR Leaderboard table.","quote":"| **Version** | **Model**     | **RTFx**   | **AMI**   | **GigaSpeech**   | **LS Clean**   | **LS Other**   | **Earnings22**   | **SPGISpech**   | **Tedlium**   | **Voxpopuli**   | ... | nemo-main  | canary-1b-flash | 1045.75 | 13.11 | 9.85 | 1.48 | 2.87 | 12.79 | 1.95 | 3.12 | 5.63 |","where_in_source":"Open ASR Leaderboard table, LibriSpeech test-clean column","fact_url":"https://yapapp.cc/facts/nvidia-canary-1b-flash-librispeech-test-clean-open-asr-1-48","source_title":"canary-1b-flash model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"n.d.","source_url":"https://huggingface.co/nvidia/canary-1b-flash/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Small, LS test-clean English","value":1.53,"low":null,"high":null,"as_written":"1.53","unit":"percent","system":"Voxtral Small","kind":"Open model","benchmark":"LibriSpeech test-clean","language":"English","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, Voxtral Small had a word error rate of 1.53 percent on LibriSpeech test-clean.","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... Voxtral Small 1.53 3.14 10.27 6.62 11.09 9.64 1.89 9.55 12.48","where_in_source":"Table 3 (short-form English), LibriSpeech test-clean column","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-small-librispeech-test-clean-1-53","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini Transcribe, LS test-clean","value":1.57,"low":null,"high":null,"as_written":"1.57","unit":"percent","system":"Voxtral Mini Transcribe","kind":"Open model","benchmark":"LibriSpeech test-clean","language":"English","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, Voxtral Mini Transcribe had a word error rate of 1.57 percent on LibriSpeech test-clean.","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... Voxtral Mini Transcribe 1.57 3.21 10.04 6.78 11.35 10.03 2.04 9.52 12.18","where_in_source":"Table 3 (short-form English), LibriSpeech test-clean column","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-transcribe-librispeech-test-clean-1-57","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Canary-Qwen 2.5B, LS test-clean English","value":1.6,"low":null,"high":null,"as_written":"1.6","unit":"percent","system":"NVIDIA Canary Qwen 2.5B","kind":"Open model","benchmark":"LibriSpeech test-clean","language":"English","condition":"Clean read speech","year":2025,"conversion":null,"claim":"NVIDIA Canary Qwen 2.5B had a word error rate of 1.60 percent on the LibriSpeech test-clean test set in the Open ASR Leaderboard table.","quote":"| **Version** | **Model**     | **RTFx**   | **Mean**   | **AMI**   | **GigaSpeech**   | **LS Clean**   | **LS Other**   | **Earnings22**   | **SPGISpech**   | **Tedlium**   | **Voxpopuli**   | ... | 2.5.0  | Canary-Qwen-2.5B | 418 | 5.63 | 10.18 | 9.41 | 1.60 | 3.10 | 10.42 | 1.90 | 2.72 | 5.66 |","where_in_source":"Open ASR Leaderboard table, LibriSpeech test-clean column","fact_url":"https://yapapp.cc/facts/nvidia-canary-qwen-2-5b-librispeech-test-clean-open-asr-1-60","source_title":"canary-qwen-2.5b model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/canary-qwen-2.5b/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"wav2vec 2.0, all labeled data","value":1.8,"low":null,"high":null,"as_written":"1.8","unit":"percent","system":"wav2vec 2.0","kind":"Open model","benchmark":"LibriSpeech test-clean","language":"English","condition":"Clean read speech","year":2020,"conversion":"The fact also gives 3.3 on test-other; that figure is a separate row. Language is taken from the test set name: LibriSpeech is English.","claim":"With all of LibriSpeech's labeled training data, wav2vec 2.0 reached word error rates of 1.8 percent on the clean test set and 3.3 percent on the harder other test set.","quote":"Experiments using all labeled data of Librispeech achieve 1.8/3.3 WER on the clean/other test sets.","where_in_source":"Abstract","fact_url":"https://yapapp.cc/facts/wav2vec2-librispeech-1-8-3-3","source_title":"wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations","source_publisher":"arXiv (NeurIPS 2020)","source_year":"2020","source_url":"https://arxiv.org/abs/2006.11477","source_doi":null,"checked":"2026-10-16"},{"row":"Whisper large-v3, LS test-clean English","value":1.84,"low":null,"high":null,"as_written":"1.84","unit":"percent","system":"Whisper large-v3","kind":"Open model","benchmark":"LibriSpeech test-clean","language":"English","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, Whisper large-v3 had a word error rate of 1.84 percent on LibriSpeech test-clean.","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... Whisper large-v3 1.84 3.66 11.60 9.58 13.14 10.88 3.15 9.88 13.07","where_in_source":"Table 3 (short-form English), LibriSpeech test-clean column","fact_url":"https://yapapp.cc/facts/voxtral-2025-whisper-large-v3-librispeech-test-clean-1-84","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini, LS test-clean English","value":1.86,"low":null,"high":null,"as_written":"1.86","unit":"percent","system":"Voxtral Mini","kind":"Open model","benchmark":"LibriSpeech test-clean","language":"English","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, Voxtral Mini had a word error rate of 1.86 percent on LibriSpeech test-clean.","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... Voxtral Mini 1.86 4.04 10.68 6.85 11.32 10.59 2.19 9.62 11.84","where_in_source":"Table 3 (short-form English), LibriSpeech test-clean column","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-librispeech-test-clean-1-86","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Small, SPGI English","value":1.89,"low":null,"high":null,"as_written":"1.89","unit":"percent","system":"Voxtral Small","kind":"Open model","benchmark":"SPGI Speech","language":"English","condition":"Specialist vocabulary","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, Voxtral Small had a word error rate of 1.89 percent on SPGI Speech.","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... Voxtral Small 1.53 3.14 10.27 6.62 11.09 9.64 1.89 9.55 12.48","where_in_source":"Table 3 (short-form English), SPGI Speech column","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-small-spgi-speech-1-89","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Canary-Qwen 2.5B, SPGI English","value":1.9,"low":null,"high":null,"as_written":"1.9","unit":"percent","system":"NVIDIA Canary Qwen 2.5B","kind":"Open model","benchmark":"SPGI Speech","language":"English","condition":"Specialist vocabulary","year":2025,"conversion":null,"claim":"NVIDIA Canary Qwen 2.5B had a word error rate of 1.90 percent on the SPGI Speech test set in the Open ASR Leaderboard table.","quote":"| **Version** | **Model**     | **RTFx**   | **Mean**   | **AMI**   | **GigaSpeech**   | **LS Clean**   | **LS Other**   | **Earnings22**   | **SPGISpech**   | **Tedlium**   | **Voxpopuli**   | ... | 2.5.0  | Canary-Qwen-2.5B | 418 | 5.63 | 10.18 | 9.41 | 1.60 | 3.10 | 10.42 | 1.90 | 2.72 | 5.66 |","where_in_source":"Open ASR Leaderboard table, SPGI Speech column","fact_url":"https://yapapp.cc/facts/nvidia-canary-qwen-2-5b-spgi-speech-open-asr-1-90","source_title":"canary-qwen-2.5b model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/canary-qwen-2.5b/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, LS test-clean English","value":1.93,"low":null,"high":null,"as_written":"1.93","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"LibriSpeech test-clean","language":"English","condition":"Clean read speech","year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 1.93 percent on the LibriSpeech test-clean test set in the Open ASR Leaderboard table.","quote":"| **Model** | **Avg WER** | **AMI** | **Earnings-22** | **GigaSpeech** | **LS test-clean** | **LS test-other** | **SPGI Speech** | **TEDLIUM-v3** | **VoxPopuli** | ... | `parakeet-tdt-0.6b-v3` | 6.34% | 11.31% | 11.42% | 9.59% | 1.93% | 3.59% | 3.97% | 2.75% | 6.14% |","where_in_source":"Open-ASR-Leaderboard table, LibriSpeech test-clean column","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-librispeech-test-clean-open-asr-1-93","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Canary 1B Flash, SPGI English","value":1.95,"low":null,"high":null,"as_written":"1.95","unit":"percent","system":"NVIDIA Canary 1B Flash","kind":"Open model","benchmark":"SPGI Speech","language":"English","condition":"Specialist vocabulary","year":null,"conversion":null,"claim":"NVIDIA Canary 1B Flash had a word error rate of 1.95 percent on the SPGI Speech test set in the Open ASR Leaderboard table.","quote":"| **Version** | **Model**     | **RTFx**   | **AMI**   | **GigaSpeech**   | **LS Clean**   | **LS Other**   | **Earnings22**   | **SPGISpech**   | **Tedlium**   | **Voxpopuli**   | ... | nemo-main  | canary-1b-flash | 1045.75 | 13.11 | 9.85 | 1.48 | 2.87 | 12.79 | 1.95 | 3.12 | 5.63 |","where_in_source":"Open ASR Leaderboard table, SPGI Speech column","fact_url":"https://yapapp.cc/facts/nvidia-canary-1b-flash-spgi-speech-open-asr-1-95","source_title":"canary-1b-flash model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"n.d.","source_url":"https://huggingface.co/nvidia/canary-1b-flash/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini Transcribe, SPGI English","value":2.04,"low":null,"high":null,"as_written":"2.04","unit":"percent","system":"Voxtral Mini Transcribe","kind":"Open model","benchmark":"SPGI Speech","language":"English","condition":"Specialist vocabulary","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, Voxtral Mini Transcribe had a word error rate of 2.04 percent on SPGI Speech.","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... Voxtral Mini Transcribe 1.57 3.21 10.04 6.78 11.35 10.03 2.04 9.52 12.18","where_in_source":"Table 3 (short-form English), SPGI Speech column","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-transcribe-spgi-speech-2-04","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Canary 1B v2, LS test-clean English","value":2.18,"low":null,"high":null,"as_written":"2.18","unit":"percent","system":"NVIDIA Canary 1B v2","kind":"Open model","benchmark":"LibriSpeech test-clean","language":"English","condition":"Clean read speech","year":2025,"conversion":null,"claim":"NVIDIA Canary 1B v2 had a word error rate of 2.18 percent on the LibriSpeech test-clean test set in the Open ASR Leaderboard table.","quote":"|  **WER ↓**   | **RTFx**   | **Mean**   | **AMI**   | **GigaSpeech**   | **LS Clean**   | **LS Other**   | **Earnings22**   | **SPGISpech**   | **Tedlium**   | **Voxpopuli**   | ... | `Canary-1b-v2` | 749 | 7.15 | 16.01 | 10.82 | 2.18 | 3.56 | 11.79 | 2.28 | 4.29 | 6.25 |","where_in_source":"Open ASR Leaderboard table, LibriSpeech test-clean column","fact_url":"https://yapapp.cc/facts/nvidia-canary-1b-v2-librispeech-test-clean-open-asr-2-18","source_title":"canary-1b-v2 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/canary-1b-v2/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini, SPGI English","value":2.19,"low":null,"high":null,"as_written":"2.19","unit":"percent","system":"Voxtral Mini","kind":"Open model","benchmark":"SPGI Speech","language":"English","condition":"Specialist vocabulary","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, Voxtral Mini had a word error rate of 2.19 percent on SPGI Speech.","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... Voxtral Mini 1.86 4.04 10.68 6.85 11.32 10.59 2.19 9.62 11.84","where_in_source":"Table 3 (short-form English), SPGI Speech column","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-spgi-speech-2-19","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Canary 1B v2, SPGI English","value":2.28,"low":null,"high":null,"as_written":"2.28","unit":"percent","system":"NVIDIA Canary 1B v2","kind":"Open model","benchmark":"SPGI Speech","language":"English","condition":"Specialist vocabulary","year":2025,"conversion":null,"claim":"NVIDIA Canary 1B v2 had a word error rate of 2.28 percent on the SPGI Speech test set in the Open ASR Leaderboard table.","quote":"|  **WER ↓**   | **RTFx**   | **Mean**   | **AMI**   | **GigaSpeech**   | **LS Clean**   | **LS Other**   | **Earnings22**   | **SPGISpech**   | **Tedlium**   | **Voxpopuli**   | ... | `Canary-1b-v2` | 749 | 7.15 | 16.01 | 10.82 | 2.18 | 3.56 | 11.79 | 2.28 | 4.29 | 6.25 |","where_in_source":"Open ASR Leaderboard table, SPGI Speech column","fact_url":"https://yapapp.cc/facts/nvidia-canary-1b-v2-spgi-speech-open-asr-2-28","source_title":"canary-1b-v2 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/canary-1b-v2/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Canary 1B v2, noise 10 dB","value":2.29,"low":null,"high":null,"as_written":"2.29","unit":"percent","system":"NVIDIA Canary 1B v2","kind":"Open model","benchmark":"LibriSpeech test-clean with MUSAN noise","language":"English","condition":"Noisy or distant","year":2025,"conversion":null,"claim":"NVIDIA Canary 1B v2 had a word error rate of 2.29 percent on LibriSpeech Clean with MUSAN noise at 10 dB SNR.","quote":"Performance across different Signal-to-Noise Ratios (SNR) using MUSAN music and noise samples \\[16] on the [LibriSpeech Clean test set](https://www.openslr.org/12). ... | **SNR (dB)**       | 100      | 10    | 5     | 0     | -5     | ... | **`Canary-1b-v2`** | 2.18% | 2.29% | 2.80% | 5.08% | 19.38% |","where_in_source":"Noise robustness table, SNR 10 dB column","fact_url":"https://yapapp.cc/facts/nvidia-canary-1b-v2-musan-snr-10-db-2-29","source_title":"canary-1b-v2 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/canary-1b-v2/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini Transcribe, FLEURS Italian","value":2.31,"low":null,"high":null,"as_written":"2.31","unit":"percent","system":"Voxtral Mini Transcribe","kind":"Open model","benchmark":"FLEURS","language":"Italian","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, Voxtral Mini Transcribe had a word error rate of 2.31 percent on FLEURS Italian.","quote":"Model ar nl en fr de hi it pt es ... Voxtral Mini Transcribe 14.64 4.89 3.61 4.22 3.54 10.32 2.31 3.57 2.75","where_in_source":"Table 4 (FLEURS), column it","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-transcribe-fleurs-it-2-31","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Canary-Qwen 2.5B, noise 10 dB","value":2.41,"low":null,"high":null,"as_written":"2.41","unit":"percent","system":"NVIDIA Canary Qwen 2.5B","kind":"Open model","benchmark":"LibriSpeech test-clean with white noise","language":"English","condition":"Noisy or distant","year":2025,"conversion":null,"claim":"NVIDIA Canary Qwen 2.5B had a word error rate of 2.41 percent on LibriSpeech test-clean with added white noise at 10 dB SNR.","quote":"WER on [Librispeech Test Clean](https://www.openslr.org/12) at different SNR (signal to noise ratio) levels of additive white noise | **Version** | **Model** | **SNR 10** | **SNR 5** | **SNR 0** | **SNR -5** | ... | 2.5.0       | Canary-Qwen-2.5B |    2.41%   |   4.08%   |   9.83%   |    30.60%  |","where_in_source":"White-noise table, SNR 10 dB column","fact_url":"https://yapapp.cc/facts/nvidia-canary-qwen-2-5b-white-noise-snr-10-db-2-41","source_title":"canary-qwen-2.5b model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/canary-qwen-2.5b/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini, FLEURS Italian","value":2.51,"low":null,"high":null,"as_written":"2.51","unit":"percent","system":"Voxtral Mini","kind":"Open model","benchmark":"FLEURS","language":"Italian","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, Voxtral Mini had a word error rate of 2.51 percent on FLEURS Italian.","quote":"Model ar nl en fr de hi it pt es ... Voxtral Mini 25.40 6.27 3.77 4.87 4.40 9.26 2.51 3.76 3.52","where_in_source":"Table 4 (FLEURS), column it","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-fleurs-it-2-51","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, Italian (our test)","value":2.52,"low":null,"high":null,"as_written":"2.52","unit":"percent","system":"Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"FLEURS","language":"Italian","condition":"Clean read speech","year":2026,"conversion":null,"claim":"In our October 2026 benchmark on a Mac with an M2 Pro, Parakeet TDT 0.6B v3 had a word error rate of 2.52 percent on 100 Italian recordings from the FLEURS test set.","quote":null,"where_in_source":null,"fact_url":"https://yapapp.cc/facts/m-fleurs-2026-10-parakeet-it-wer","source_title":null,"source_publisher":null,"source_year":null,"source_url":null,"source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v3-turbo, Spanish (our test)","value":2.53,"low":null,"high":null,"as_written":"2.53","unit":"percent","system":"Whisper large-v3-turbo","kind":"Open model","benchmark":"FLEURS","language":"Spanish","condition":"Clean read speech","year":2026,"conversion":null,"claim":"In our October 2026 benchmark on a Mac with an M2 Pro, Whisper large-v3-turbo had a word error rate of 2.53 percent on 100 Spanish recordings from the FLEURS test set.","quote":null,"where_in_source":null,"fact_url":"https://yapapp.cc/facts/m-fleurs-2026-10-whisper-es-wer","source_title":null,"source_publisher":null,"source_year":null,"source_url":null,"source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Small, FLEURS Italian","value":2.62,"low":null,"high":null,"as_written":"2.62","unit":"percent","system":"Voxtral Small","kind":"Open model","benchmark":"FLEURS","language":"Italian","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, Voxtral Small had a word error rate of 2.62 percent on FLEURS Italian.","quote":"Model ar nl en fr de hi it pt es ... Voxtral Small 13.44 4.94 3.35 4.03 3.38 7.69 2.62 3.79 2.72","where_in_source":"Table 4 (FLEURS), column it","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-small-fleurs-it-2-62","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Canary 1B Flash, MLS Spanish","value":2.69,"low":null,"high":null,"as_written":"2.69","unit":"percent","system":"NVIDIA Canary 1B Flash","kind":"Open model","benchmark":"Multilingual LibriSpeech","language":"Spanish","condition":"Clean read speech","year":null,"conversion":null,"claim":"NVIDIA Canary 1B Flash had a word error rate of 2.69 percent on Multilingual LibriSpeech Spanish.","quote":"WER on [MLS](https://huggingface.co/datasets/facebook/multilingual_librispeech) test set: | **Version** | **Model**  | **De**   | **Es**   | **Fr**   | ... | nemo-main   | canary-1b-flash | 4.36 | 2.69 | 4.47 |","where_in_source":"WER table, es column","fact_url":"https://yapapp.cc/facts/nvidia-canary-1b-flash-mls-es-2-69","source_title":"canary-1b-flash model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"n.d.","source_url":"https://huggingface.co/nvidia/canary-1b-flash/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v2, LibriSpeech clean","value":2.7,"low":null,"high":null,"as_written":"2.7","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"LibriSpeech test-clean","language":"English","condition":"Clean read speech","year":2022,"conversion":"The source's LibriSpeech Clean column is the test-clean split. The language is English because LibriSpeech is an English audiobook corpus. The figure is the second column of the row, after the wav2vec 2.0 figure.","claim":"On the LibriSpeech Clean test set, Whisper Large V2 had a word error rate of 2.7 percent.","quote":"LibriSpeech Clean 2.7 2.7 0.0 Artie 24.5 6.2 74.7","where_in_source":"Table 2, effective robustness across datasets","fact_url":"https://yapapp.cc/facts/whisper-librispeech-clean-wer-2-7","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision","source_publisher":"arXiv (eess.AS)","source_year":"2022","source_url":"https://arxiv.org/abs/2212.04356","source_doi":null,"checked":"2026-10-16"},{"row":"Whisper large-v3, FLEURS Italian","value":2.71,"low":null,"high":null,"as_written":"2.71","unit":"percent","system":"Whisper large-v3","kind":"Open model","benchmark":"FLEURS","language":"Italian","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, Whisper large-v3 had a word error rate of 2.71 percent on FLEURS Italian.","quote":"Model ar nl en fr de hi it pt es ... Whisper large-v3 15.44 5.87 4.00 5.55 5.46 28.87 2.71 3.90 2.81","where_in_source":"Table 4 (FLEURS), column it","fact_url":"https://yapapp.cc/facts/voxtral-2025-whisper-large-v3-fleurs-it-2-71","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Canary-Qwen 2.5B, TED-LIUM English","value":2.72,"low":null,"high":null,"as_written":"2.72","unit":"percent","system":"NVIDIA Canary Qwen 2.5B","kind":"Open model","benchmark":"TED-LIUM","language":"English","condition":"Mixed benchmark","year":2025,"conversion":null,"claim":"NVIDIA Canary Qwen 2.5B had a word error rate of 2.72 percent on the TED-LIUM test set in the Open ASR Leaderboard table.","quote":"| **Version** | **Model**     | **RTFx**   | **Mean**   | **AMI**   | **GigaSpeech**   | **LS Clean**   | **LS Other**   | **Earnings22**   | **SPGISpech**   | **Tedlium**   | **Voxpopuli**   | ... | 2.5.0  | Canary-Qwen-2.5B | 418 | 5.63 | 10.18 | 9.41 | 1.60 | 3.10 | 10.42 | 1.90 | 2.72 | 5.66 |","where_in_source":"Open ASR Leaderboard table, TED-LIUM column","fact_url":"https://yapapp.cc/facts/nvidia-canary-qwen-2-5b-ted-lium-open-asr-2-72","source_title":"canary-qwen-2.5b model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/canary-qwen-2.5b/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Small, FLEURS Spanish","value":2.72,"low":null,"high":null,"as_written":"2.72","unit":"percent","system":"Voxtral Small","kind":"Open model","benchmark":"FLEURS","language":"Spanish","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, Voxtral Small had a word error rate of 2.72 percent on FLEURS Spanish.","quote":"Model ar nl en fr de hi it pt es ... Voxtral Small 13.44 4.94 3.35 4.03 3.38 7.69 2.62 3.79 2.72","where_in_source":"Table 4 (FLEURS), column es","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-small-fleurs-es-2-72","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v3-turbo, Italian (our test)","value":2.73,"low":null,"high":null,"as_written":"2.73","unit":"percent","system":"Whisper large-v3-turbo","kind":"Open model","benchmark":"FLEURS","language":"Italian","condition":"Clean read speech","year":2026,"conversion":null,"claim":"In our October 2026 benchmark on a Mac with an M2 Pro, Whisper large-v3-turbo had a word error rate of 2.73 percent on 100 Italian recordings from the FLEURS test set.","quote":null,"where_in_source":null,"fact_url":"https://yapapp.cc/facts/m-fleurs-2026-10-whisper-it-wer","source_title":null,"source_publisher":null,"source_year":null,"source_url":null,"source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, TED-LIUM English","value":2.75,"low":null,"high":null,"as_written":"2.75","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"TED-LIUM","language":"English","condition":"Mixed benchmark","year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 2.75 percent on the TED-LIUM test set in the Open ASR Leaderboard table.","quote":"| **Model** | **Avg WER** | **AMI** | **Earnings-22** | **GigaSpeech** | **LS test-clean** | **LS test-other** | **SPGI Speech** | **TEDLIUM-v3** | **VoxPopuli** | ... | `parakeet-tdt-0.6b-v3` | 6.34% | 11.31% | 11.42% | 9.59% | 1.93% | 3.59% | 3.97% | 2.75% | 6.14% |","where_in_source":"Open-ASR-Leaderboard table, TED-LIUM column","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-ted-lium-open-asr-2-75","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini Transcribe, FLEURS Spanish","value":2.75,"low":null,"high":null,"as_written":"2.75","unit":"percent","system":"Voxtral Mini Transcribe","kind":"Open model","benchmark":"FLEURS","language":"Spanish","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, Voxtral Mini Transcribe had a word error rate of 2.75 percent on FLEURS Spanish.","quote":"Model ar nl en fr de hi it pt es ... Voxtral Mini Transcribe 14.64 4.89 3.61 4.22 3.54 10.32 2.31 3.57 2.75","where_in_source":"Table 4 (FLEURS), column es","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-transcribe-fleurs-es-2-75","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, Spanish (our test)","value":2.79,"low":null,"high":null,"as_written":"2.79","unit":"percent","system":"Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"FLEURS","language":"Spanish","condition":"Clean read speech","year":2026,"conversion":null,"claim":"In our October 2026 benchmark on a Mac with an M2 Pro, Parakeet TDT 0.6B v3 had a word error rate of 2.79 percent on 100 Spanish recordings from the FLEURS test set.","quote":null,"where_in_source":null,"fact_url":"https://yapapp.cc/facts/m-fleurs-2026-10-parakeet-es-wer","source_title":null,"source_publisher":null,"source_year":null,"source_url":null,"source_doi":null,"checked":"2026-10-19"},{"row":"Canary 1B v2, noise 5 dB","value":2.8,"low":null,"high":null,"as_written":"2.8","unit":"percent","system":"NVIDIA Canary 1B v2","kind":"Open model","benchmark":"LibriSpeech test-clean with MUSAN noise","language":"English","condition":"Noisy or distant","year":2025,"conversion":null,"claim":"NVIDIA Canary 1B v2 had a word error rate of 2.80 percent on LibriSpeech Clean with MUSAN noise at 5 dB SNR.","quote":"Performance across different Signal-to-Noise Ratios (SNR) using MUSAN music and noise samples \\[16] on the [LibriSpeech Clean test set](https://www.openslr.org/12). ... | **SNR (dB)**       | 100      | 10    | 5     | 0     | -5     | ... | **`Canary-1b-v2`** | 2.18% | 2.29% | 2.80% | 5.08% | 19.38% |","where_in_source":"Noise robustness table, SNR 5 dB column","fact_url":"https://yapapp.cc/facts/nvidia-canary-1b-v2-musan-snr-5-db-2-80","source_title":"canary-1b-v2 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/canary-1b-v2/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v3, FLEURS Spanish","value":2.81,"low":null,"high":null,"as_written":"2.81","unit":"percent","system":"Whisper large-v3","kind":"Open model","benchmark":"FLEURS","language":"Spanish","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, Whisper large-v3 had a word error rate of 2.81 percent on FLEURS Spanish.","quote":"Model ar nl en fr de hi it pt es ... Whisper large-v3 15.44 5.87 4.00 5.55 5.46 28.87 2.71 3.90 2.81","where_in_source":"Table 4 (FLEURS), column es","fact_url":"https://yapapp.cc/facts/voxtral-2025-whisper-large-v3-fleurs-es-2-81","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Canary 1B Flash, LS test-other English","value":2.87,"low":null,"high":null,"as_written":"2.87","unit":"percent","system":"NVIDIA Canary 1B Flash","kind":"Open model","benchmark":"LibriSpeech test-other","language":"English","condition":"Mixed benchmark","year":null,"conversion":null,"claim":"NVIDIA Canary 1B Flash had a word error rate of 2.87 percent on the LibriSpeech test-other test set in the Open ASR Leaderboard table.","quote":"| **Version** | **Model**     | **RTFx**   | **AMI**   | **GigaSpeech**   | **LS Clean**   | **LS Other**   | **Earnings22**   | **SPGISpech**   | **Tedlium**   | **Voxpopuli**   | ... | nemo-main  | canary-1b-flash | 1045.75 | 13.11 | 9.85 | 1.48 | 2.87 | 12.79 | 1.95 | 3.12 | 5.63 |","where_in_source":"Open ASR Leaderboard table, LibriSpeech test-other column","fact_url":"https://yapapp.cc/facts/nvidia-canary-1b-flash-librispeech-test-other-open-asr-2-87","source_title":"canary-1b-flash model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"n.d.","source_url":"https://huggingface.co/nvidia/canary-1b-flash/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Kuhn et al. 2024, Whisper large-v2","value":2.9,"low":null,"high":null,"as_written":"2.9","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"University lecture recordings","language":"English","condition":null,"year":2024,"conversion":"Kind is inferred: the source lists Whisper (large-v2) as a model beside the commercial services, and Whisper is an open model.","claim":"In the same 2024 study, Whisper large-v2 had an average word error rate of 2.9 percent on the English lecture recordings.","quote":"Speechmatics 3.3 3.6 4.6 8.0 Tencent 4.6 4.1 7.3 - Whisper (large-v2) 2.9 3.3 3.3 5.0","where_in_source":"Table 4","fact_url":"https://yapapp.cc/facts/kuhn-2024-whisper-large-v2-english-average-wer-2-9","source_title":"Measuring the Accuracy of Automatic Speech Recognition Solutions","source_publisher":"arXiv (preprint)","source_year":"2024","source_url":"https://arxiv.org/pdf/2408.16287","source_doi":null,"checked":"2026-10-17"},{"row":"Parakeet TDT v3, FLEURS Italian","value":3,"low":null,"high":null,"as_written":"3","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"FLEURS","language":"Italian","condition":"Clean read speech","year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 3.00 percent on the FLEURS Italian test.","quote":"| Language | Fleurs | MLS | CoVoST | ... | **it** | 3.00% | 10.08% | 3.69% |","where_in_source":"Multilingual ASR table, Fleurs column, it row","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-fleurs-it-3-00","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v2, Spanish","value":3,"low":null,"high":null,"as_written":"3","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Spanish","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the Fleurs benchmark, the Whisper large-v2 model had a word error rate of 3.0 percent for Spanish.","quote":"Model German Greek English Spanish Estonian Persian Finnish Tagalog French Galician Gujarati Hausa Hebrew Hindi ... Whisper large-v2 4.5 12.5 4.2 3.0 21.9 32.9 9.7 13.8 8.3 15.4 102.7 88.9 27.1 21.5","where_in_source":"Table 13 (Fleurs)","fact_url":"https://yapapp.cc/facts/whisper-fleurs-spanish-3-0","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-16"},{"row":"Canary 1B, MLS English","value":3.06,"low":null,"high":null,"as_written":"3.06","unit":"percent","system":"NVIDIA Canary 1B","kind":"Open model","benchmark":"Multilingual LibriSpeech","language":"English","condition":"Clean read speech","year":null,"conversion":null,"claim":"NVIDIA Canary 1B had a word error rate of 3.06 percent on Multilingual LibriSpeech English.","quote":"WER on [MLS](https://huggingface.co/datasets/facebook/multilingual_librispeech) test set: | **Version** | **Model**     | **En**   | **De**   | **Es**   | **Fr**   | ... | 1.23.0  | canary-1b | 3.06 | 4.19 | 3.15 | 4.12 |","where_in_source":"WER table, en column","fact_url":"https://yapapp.cc/facts/nvidia-canary-1b-mls-en-3-06","source_title":"canary-1b model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"n.d.","source_url":"https://huggingface.co/nvidia/canary-1b/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Canary-Qwen 2.5B, LS test-other English","value":3.1,"low":null,"high":null,"as_written":"3.1","unit":"percent","system":"NVIDIA Canary Qwen 2.5B","kind":"Open model","benchmark":"LibriSpeech test-other","language":"English","condition":"Mixed benchmark","year":2025,"conversion":null,"claim":"NVIDIA Canary Qwen 2.5B had a word error rate of 3.10 percent on the LibriSpeech test-other test set in the Open ASR Leaderboard table.","quote":"| **Version** | **Model**     | **RTFx**   | **Mean**   | **AMI**   | **GigaSpeech**   | **LS Clean**   | **LS Other**   | **Earnings22**   | **SPGISpech**   | **Tedlium**   | **Voxpopuli**   | ... | 2.5.0  | Canary-Qwen-2.5B | 418 | 5.63 | 10.18 | 9.41 | 1.60 | 3.10 | 10.42 | 1.90 | 2.72 | 5.66 |","where_in_source":"Open ASR Leaderboard table, LibriSpeech test-other column","fact_url":"https://yapapp.cc/facts/nvidia-canary-qwen-2-5b-librispeech-test-other-open-asr-3-10","source_title":"canary-qwen-2.5b model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/canary-qwen-2.5b/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Canary 1B Flash, TED-LIUM English","value":3.12,"low":null,"high":null,"as_written":"3.12","unit":"percent","system":"NVIDIA Canary 1B Flash","kind":"Open model","benchmark":"TED-LIUM","language":"English","condition":"Mixed benchmark","year":null,"conversion":null,"claim":"NVIDIA Canary 1B Flash had a word error rate of 3.12 percent on the TED-LIUM test set in the Open ASR Leaderboard table.","quote":"| **Version** | **Model**     | **RTFx**   | **AMI**   | **GigaSpeech**   | **LS Clean**   | **LS Other**   | **Earnings22**   | **SPGISpech**   | **Tedlium**   | **Voxpopuli**   | ... | nemo-main  | canary-1b-flash | 1045.75 | 13.11 | 9.85 | 1.48 | 2.87 | 12.79 | 1.95 | 3.12 | 5.63 |","where_in_source":"Open ASR Leaderboard table, TED-LIUM column","fact_url":"https://yapapp.cc/facts/nvidia-canary-1b-flash-ted-lium-open-asr-3-12","source_title":"canary-1b-flash model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"n.d.","source_url":"https://huggingface.co/nvidia/canary-1b-flash/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Small, LS test-other English","value":3.14,"low":null,"high":null,"as_written":"3.14","unit":"percent","system":"Voxtral Small","kind":"Open model","benchmark":"LibriSpeech test-other","language":"English","condition":"Mixed benchmark","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, Voxtral Small had a word error rate of 3.14 percent on LibriSpeech test-other.","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... Voxtral Small 1.53 3.14 10.27 6.62 11.09 9.64 1.89 9.55 12.48","where_in_source":"Table 3 (short-form English), LibriSpeech test-other column","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-small-librispeech-test-other-3-14","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Canary 1B, MLS Spanish","value":3.15,"low":null,"high":null,"as_written":"3.15","unit":"percent","system":"NVIDIA Canary 1B","kind":"Open model","benchmark":"Multilingual LibriSpeech","language":"Spanish","condition":"Clean read speech","year":null,"conversion":null,"claim":"NVIDIA Canary 1B had a word error rate of 3.15 percent on Multilingual LibriSpeech Spanish.","quote":"WER on [MLS](https://huggingface.co/datasets/facebook/multilingual_librispeech) test set: | **Version** | **Model**     | **En**   | **De**   | **Es**   | **Fr**   | ... | 1.23.0  | canary-1b | 3.06 | 4.19 | 3.15 | 4.12 |","where_in_source":"WER table, es column","fact_url":"https://yapapp.cc/facts/nvidia-canary-1b-mls-es-3-15","source_title":"canary-1b model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"n.d.","source_url":"https://huggingface.co/nvidia/canary-1b/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v3, SPGI English","value":3.15,"low":null,"high":null,"as_written":"3.15","unit":"percent","system":"Whisper large-v3","kind":"Open model","benchmark":"SPGI Speech","language":"English","condition":"Specialist vocabulary","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, Whisper large-v3 had a word error rate of 3.15 percent on SPGI Speech.","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... Whisper large-v3 1.84 3.66 11.60 9.58 13.14 10.88 3.15 9.88 13.07","where_in_source":"Table 3 (short-form English), SPGI Speech column","fact_url":"https://yapapp.cc/facts/voxtral-2025-whisper-large-v3-spgi-speech-3-15","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini Transcribe, LS test-other","value":3.21,"low":null,"high":null,"as_written":"3.21","unit":"percent","system":"Voxtral Mini Transcribe","kind":"Open model","benchmark":"LibriSpeech test-other","language":"English","condition":"Mixed benchmark","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, Voxtral Mini Transcribe had a word error rate of 3.21 percent on LibriSpeech test-other.","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... Voxtral Mini Transcribe 1.57 3.21 10.04 6.78 11.35 10.03 2.04 9.52 12.18","where_in_source":"Table 3 (short-form English), LibriSpeech test-other column","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-transcribe-librispeech-test-other-3-21","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v3-turbo, Portuguese (our test)","value":3.28,"low":null,"high":null,"as_written":"3.28","unit":"percent","system":"Whisper large-v3-turbo","kind":"Open model","benchmark":"FLEURS","language":"Portuguese","condition":"Clean read speech","year":2026,"conversion":null,"claim":"In our October 2026 benchmark on a Mac with an M2 Pro, Whisper large-v3-turbo had a word error rate of 3.28 percent on 100 Portuguese recordings from the FLEURS test set.","quote":null,"where_in_source":null,"fact_url":"https://yapapp.cc/facts/m-fleurs-2026-10-whisper-pt-wer","source_title":null,"source_publisher":null,"source_year":null,"source_url":null,"source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Small, CV Spanish","value":3.31,"low":null,"high":null,"as_written":"3.31","unit":"percent","system":"Voxtral Small","kind":"Open model","benchmark":"Common Voice","language":"Spanish","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, Voxtral Small had a word error rate of 3.31 percent on Common Voice Spanish.","quote":"Model ar nl en fr de hi it pt es ... Voxtral Small 61.97 3.98 8.58 6.18 3.74 9.01 3.96 6.43 3.31","where_in_source":"Table 5 (Common Voice), column es","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-small-common-voice-es-3-31","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Small, FLEURS English","value":3.35,"low":null,"high":null,"as_written":"3.35","unit":"percent","system":"Voxtral Small","kind":"Open model","benchmark":"FLEURS","language":"English","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, Voxtral Small had a word error rate of 3.35 percent on FLEURS English.","quote":"Model ar nl en fr de hi it pt es ... Voxtral Small 13.44 4.94 3.35 4.03 3.38 7.69 2.62 3.79 2.72","where_in_source":"Table 4 (FLEURS), column en","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-small-fleurs-en-3-35","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Small, FLEURS German","value":3.38,"low":null,"high":null,"as_written":"3.38","unit":"percent","system":"Voxtral Small","kind":"Open model","benchmark":"FLEURS","language":"German","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, Voxtral Small had a word error rate of 3.38 percent on FLEURS German.","quote":"Model ar nl en fr de hi it pt es ... Voxtral Small 13.44 4.94 3.35 4.03 3.38 7.69 2.62 3.79 2.72","where_in_source":"Table 4 (FLEURS), column de","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-small-fleurs-de-3-38","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, CoVoST 2 Spanish","value":3.41,"low":null,"high":null,"as_written":"3.41","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"CoVoST 2","language":"Spanish","condition":null,"year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 3.41 percent on the CoVoST 2 Spanish test.","quote":"| Language | Fleurs | MLS | CoVoST | ... | **es** | 3.45% | 4.39% | 3.41% |","where_in_source":"Multilingual ASR table, CoVoST column, es row","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-covost-2-es-3-41","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, FLEURS Spanish","value":3.45,"low":null,"high":null,"as_written":"3.45","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"FLEURS","language":"Spanish","condition":"Clean read speech","year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 3.45 percent on the FLEURS Spanish test.","quote":"| Language | Fleurs | MLS | CoVoST | ... | **es** | 3.45% | 4.39% | 3.41% |","where_in_source":"Multilingual ASR table, Fleurs column, es row","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-fleurs-es-3-45","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini, FLEURS Spanish","value":3.52,"low":null,"high":null,"as_written":"3.52","unit":"percent","system":"Voxtral Mini","kind":"Open model","benchmark":"FLEURS","language":"Spanish","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, Voxtral Mini had a word error rate of 3.52 percent on FLEURS Spanish.","quote":"Model ar nl en fr de hi it pt es ... Voxtral Mini 25.40 6.27 3.77 4.87 4.40 9.26 2.51 3.76 3.52","where_in_source":"Table 4 (FLEURS), column es","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-fleurs-es-3-52","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini Transcribe, FLEURS German","value":3.54,"low":null,"high":null,"as_written":"3.54","unit":"percent","system":"Voxtral Mini Transcribe","kind":"Open model","benchmark":"FLEURS","language":"German","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, Voxtral Mini Transcribe had a word error rate of 3.54 percent on FLEURS German.","quote":"Model ar nl en fr de hi it pt es ... Voxtral Mini Transcribe 14.64 4.89 3.61 4.22 3.54 10.32 2.31 3.57 2.75","where_in_source":"Table 4 (FLEURS), column de","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-transcribe-fleurs-de-3-54","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Canary 1B v2, LS test-other English","value":3.56,"low":null,"high":null,"as_written":"3.56","unit":"percent","system":"NVIDIA Canary 1B v2","kind":"Open model","benchmark":"LibriSpeech test-other","language":"English","condition":"Mixed benchmark","year":2025,"conversion":null,"claim":"NVIDIA Canary 1B v2 had a word error rate of 3.56 percent on the LibriSpeech test-other test set in the Open ASR Leaderboard table.","quote":"|  **WER ↓**   | **RTFx**   | **Mean**   | **AMI**   | **GigaSpeech**   | **LS Clean**   | **LS Other**   | **Earnings22**   | **SPGISpech**   | **Tedlium**   | **Voxpopuli**   | ... | `Canary-1b-v2` | 749 | 7.15 | 16.01 | 10.82 | 2.18 | 3.56 | 11.79 | 2.28 | 4.29 | 6.25 |","where_in_source":"Open ASR Leaderboard table, LibriSpeech test-other column","fact_url":"https://yapapp.cc/facts/nvidia-canary-1b-v2-librispeech-test-other-open-asr-3-56","source_title":"canary-1b-v2 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/canary-1b-v2/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini Transcribe, FLEURS","value":3.57,"low":null,"high":null,"as_written":"3.57","unit":"percent","system":"Voxtral Mini Transcribe","kind":"Open model","benchmark":"FLEURS","language":"Portuguese","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, Voxtral Mini Transcribe had a word error rate of 3.57 percent on FLEURS Portuguese.","quote":"Model ar nl en fr de hi it pt es ... Voxtral Mini Transcribe 14.64 4.89 3.61 4.22 3.54 10.32 2.31 3.57 2.75","where_in_source":"Table 4 (FLEURS), column pt","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-transcribe-fleurs-pt-3-57","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, LS test-other English","value":3.59,"low":null,"high":null,"as_written":"3.59","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"LibriSpeech test-other","language":"English","condition":"Mixed benchmark","year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 3.59 percent on the LibriSpeech test-other test set in the Open ASR Leaderboard table.","quote":"| **Model** | **Avg WER** | **AMI** | **Earnings-22** | **GigaSpeech** | **LS test-clean** | **LS test-other** | **SPGI Speech** | **TEDLIUM-v3** | **VoxPopuli** | ... | `parakeet-tdt-0.6b-v3` | 6.34% | 11.31% | 11.42% | 9.59% | 1.93% | 3.59% | 3.97% | 2.75% | 6.14% |","where_in_source":"Open-ASR-Leaderboard table, LibriSpeech test-other column","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-librispeech-test-other-open-asr-3-59","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini Transcribe, FLEURS English","value":3.61,"low":null,"high":null,"as_written":"3.61","unit":"percent","system":"Voxtral Mini Transcribe","kind":"Open model","benchmark":"FLEURS","language":"English","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, Voxtral Mini Transcribe had a word error rate of 3.61 percent on FLEURS English.","quote":"Model ar nl en fr de hi it pt es ... Voxtral Mini Transcribe 14.64 4.89 3.61 4.22 3.54 10.32 2.31 3.57 2.75","where_in_source":"Table 4 (FLEURS), column en","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-transcribe-fleurs-en-3-61","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, Portuguese (our test)","value":3.62,"low":null,"high":null,"as_written":"3.62","unit":"percent","system":"Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"FLEURS","language":"Portuguese","condition":"Clean read speech","year":2026,"conversion":null,"claim":"In our October 2026 benchmark on a Mac with an M2 Pro, Parakeet TDT 0.6B v3 had a word error rate of 3.62 percent on 100 Portuguese recordings from the FLEURS test set.","quote":null,"where_in_source":null,"fact_url":"https://yapapp.cc/facts/m-fleurs-2026-10-parakeet-pt-wer","source_title":null,"source_publisher":null,"source_year":null,"source_url":null,"source_doi":null,"checked":"2026-10-19"},{"row":"Canary 1B Flash, CV Spanish","value":3.62,"low":null,"high":null,"as_written":"3.62","unit":"percent","system":"NVIDIA Canary 1B Flash","kind":"Open model","benchmark":"Common Voice","language":"Spanish","condition":null,"year":null,"conversion":null,"claim":"NVIDIA Canary 1B Flash had a word error rate of 3.62 percent on Common Voice Spanish.","quote":"WER on [MCV-16.1](https://commonvoice.mozilla.org/en/datasets) test set: | **Version** | **Model**  |  **En**   | **De**   | **Es**   | **Fr**   | ... | nemo-main   | canary-1b-flash | 6.99 | 4.09 | 3.62 | 6.15 |","where_in_source":"WER table, es column","fact_url":"https://yapapp.cc/facts/nvidia-canary-1b-flash-mcv-es-3-62","source_title":"canary-1b-flash model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"n.d.","source_url":"https://huggingface.co/nvidia/canary-1b-flash/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Small, MLS Spanish","value":3.62,"low":null,"high":null,"as_written":"3.62","unit":"percent","system":"Voxtral Small","kind":"Open model","benchmark":"Multilingual LibriSpeech","language":"Spanish","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 6, Voxtral Small had a word error rate of 3.62 percent on Multilingual LibriSpeech Spanish.","quote":"Model nl fr de it pt es ... Voxtral Small 9.43 3.73 5.57 8.44 5.85 3.62","where_in_source":"Table 6 (Multilingual LibriSpeech), column es","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-small-multilingual-librispeech-es-3-62","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v3, LS test-other English","value":3.66,"low":null,"high":null,"as_written":"3.66","unit":"percent","system":"Whisper large-v3","kind":"Open model","benchmark":"LibriSpeech test-other","language":"English","condition":"Mixed benchmark","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, Whisper large-v3 had a word error rate of 3.66 percent on LibriSpeech test-other.","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... Whisper large-v3 1.84 3.66 11.60 9.58 13.14 10.88 3.15 9.88 13.07","where_in_source":"Table 3 (short-form English), LibriSpeech test-other column","fact_url":"https://yapapp.cc/facts/voxtral-2025-whisper-large-v3-librispeech-test-other-3-66","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, CoVoST 2 Italian","value":3.69,"low":null,"high":null,"as_written":"3.69","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"CoVoST 2","language":"Italian","condition":null,"year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 3.69 percent on the CoVoST 2 Italian test.","quote":"| Language | Fleurs | MLS | CoVoST | ... | **it** | 3.00% | 10.08% | 3.69% |","where_in_source":"Multilingual ASR table, CoVoST column, it row","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-covost-2-it-3-69","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Small, MLS French","value":3.73,"low":null,"high":null,"as_written":"3.73","unit":"percent","system":"Voxtral Small","kind":"Open model","benchmark":"Multilingual LibriSpeech","language":"French","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 6, Voxtral Small had a word error rate of 3.73 percent on Multilingual LibriSpeech French.","quote":"Model nl fr de it pt es ... Voxtral Small 9.43 3.73 5.57 8.44 5.85 3.62","where_in_source":"Table 6 (Multilingual LibriSpeech), column fr","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-small-multilingual-librispeech-fr-3-73","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Small, CV German","value":3.74,"low":null,"high":null,"as_written":"3.74","unit":"percent","system":"Voxtral Small","kind":"Open model","benchmark":"Common Voice","language":"German","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, Voxtral Small had a word error rate of 3.74 percent on Common Voice German.","quote":"Model ar nl en fr de hi it pt es ... Voxtral Small 61.97 3.98 8.58 6.18 3.74 9.01 3.96 6.43 3.31","where_in_source":"Table 5 (Common Voice), column de","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-small-common-voice-de-3-74","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini, FLEURS Portuguese","value":3.76,"low":null,"high":null,"as_written":"3.76","unit":"percent","system":"Voxtral Mini","kind":"Open model","benchmark":"FLEURS","language":"Portuguese","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, Voxtral Mini had a word error rate of 3.76 percent on FLEURS Portuguese.","quote":"Model ar nl en fr de hi it pt es ... Voxtral Mini 25.40 6.27 3.77 4.87 4.40 9.26 2.51 3.76 3.52","where_in_source":"Table 4 (FLEURS), column pt","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-fleurs-pt-3-76","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini, FLEURS English","value":3.77,"low":null,"high":null,"as_written":"3.77","unit":"percent","system":"Voxtral Mini","kind":"Open model","benchmark":"FLEURS","language":"English","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, Voxtral Mini had a word error rate of 3.77 percent on FLEURS English.","quote":"Model ar nl en fr de hi it pt es ... Voxtral Mini 25.40 6.27 3.77 4.87 4.40 9.26 2.51 3.76 3.52","where_in_source":"Table 4 (FLEURS), column en","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-fleurs-en-3-77","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Small, FLEURS Portuguese","value":3.79,"low":null,"high":null,"as_written":"3.79","unit":"percent","system":"Voxtral Small","kind":"Open model","benchmark":"FLEURS","language":"Portuguese","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, Voxtral Small had a word error rate of 3.79 percent on FLEURS Portuguese.","quote":"Model ar nl en fr de hi it pt es ... Voxtral Small 13.44 4.94 3.35 4.03 3.38 7.69 2.62 3.79 2.72","where_in_source":"Table 4 (FLEURS), column pt","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-small-fleurs-pt-3-79","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini Transcribe, MLS Spanish","value":3.87,"low":null,"high":null,"as_written":"3.87","unit":"percent","system":"Voxtral Mini Transcribe","kind":"Open model","benchmark":"Multilingual LibriSpeech","language":"Spanish","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 6, Voxtral Mini Transcribe had a word error rate of 3.87 percent on Multilingual LibriSpeech Spanish.","quote":"Model nl fr de it pt es ... Voxtral Mini Transcribe 9.63 4.14 5.64 9.28 5.17 3.87","where_in_source":"Table 6 (Multilingual LibriSpeech), column es","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-transcribe-multilingual-librispeech-es-3-87","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v3, MLS Spanish","value":3.89,"low":null,"high":null,"as_written":"3.89","unit":"percent","system":"Whisper large-v3","kind":"Open model","benchmark":"Multilingual LibriSpeech","language":"Spanish","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 6, Whisper large-v3 had a word error rate of 3.89 percent on Multilingual LibriSpeech Spanish.","quote":"Model nl fr de it pt es ... Whisper large-v3 9.19 5.09 5.72 9.78 7.03 3.89","where_in_source":"Table 6 (Multilingual LibriSpeech), column es","fact_url":"https://yapapp.cc/facts/voxtral-2025-whisper-large-v3-multilingual-librispeech-es-3-89","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v3, FLEURS Portuguese","value":3.9,"low":null,"high":null,"as_written":"3.9","unit":"percent","system":"Whisper large-v3","kind":"Open model","benchmark":"FLEURS","language":"Portuguese","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, Whisper large-v3 had a word error rate of 3.90 percent on FLEURS Portuguese.","quote":"Model ar nl en fr de hi it pt es ... Whisper large-v3 15.44 5.87 4.00 5.55 5.46 28.87 2.71 3.90 2.81","where_in_source":"Table 4 (FLEURS), column pt","fact_url":"https://yapapp.cc/facts/voxtral-2025-whisper-large-v3-fleurs-pt-3-90","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v3-turbo, English (our test)","value":3.92,"low":null,"high":null,"as_written":"3.92","unit":"percent","system":"Whisper large-v3-turbo","kind":"Open model","benchmark":"FLEURS","language":"English","condition":"Clean read speech","year":2026,"conversion":null,"claim":"In our October 2026 benchmark on a Mac with an M2 Pro, Whisper large-v3-turbo had a word error rate of 3.92 percent on 100 English recordings from the FLEURS test set.","quote":null,"where_in_source":null,"fact_url":"https://yapapp.cc/facts/m-fleurs-2026-10-whisper-en-wer","source_title":null,"source_publisher":null,"source_year":null,"source_url":null,"source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, CoVoST 2 Portuguese","value":3.96,"low":null,"high":null,"as_written":"3.96","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"CoVoST 2","language":"Portuguese","condition":null,"year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 3.96 percent on the CoVoST 2 Portuguese test.","quote":"| Language | Fleurs | MLS | CoVoST | ... | **pt** | 4.76% | 7.50% | 3.96% |","where_in_source":"Multilingual ASR table, CoVoST column, pt row","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-covost-2-pt-3-96","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini Transcribe, CV Spanish","value":3.96,"low":null,"high":null,"as_written":"3.96","unit":"percent","system":"Voxtral Mini Transcribe","kind":"Open model","benchmark":"Common Voice","language":"Spanish","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, Voxtral Mini Transcribe had a word error rate of 3.96 percent on Common Voice Spanish.","quote":"Model ar nl en fr de hi it pt es ... Voxtral Mini Transcribe 62.01 4.71 8.25 7.29 4.85 10.42 4.37 6.70 3.96","where_in_source":"Table 5 (Common Voice), column es","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-transcribe-common-voice-es-3-96","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, SPGI English","value":3.97,"low":null,"high":null,"as_written":"3.97","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"SPGI Speech","language":"English","condition":"Specialist vocabulary","year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 3.97 percent on the SPGI Speech test set in the Open ASR Leaderboard table.","quote":"| **Model** | **Avg WER** | **AMI** | **Earnings-22** | **GigaSpeech** | **LS test-clean** | **LS test-other** | **SPGI Speech** | **TEDLIUM-v3** | **VoxPopuli** | ... | `parakeet-tdt-0.6b-v3` | 6.34% | 11.31% | 11.42% | 9.59% | 1.93% | 3.59% | 3.97% | 2.75% | 6.14% |","where_in_source":"Open-ASR-Leaderboard table, SPGI Speech column","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-spgi-speech-open-asr-3-97","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Small, CV Dutch","value":3.98,"low":null,"high":null,"as_written":"3.98","unit":"percent","system":"Voxtral Small","kind":"Open model","benchmark":"Common Voice","language":"Dutch","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, Voxtral Small had a word error rate of 3.98 percent on Common Voice Dutch.","quote":"Model ar nl en fr de hi it pt es ... Voxtral Small 61.97 3.98 8.58 6.18 3.74 9.01 3.96 6.43 3.31","where_in_source":"Table 5 (Common Voice), column nl","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-small-common-voice-nl-3-98","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Canary 1B, CV Spanish","value":3.99,"low":null,"high":null,"as_written":"3.99","unit":"percent","system":"NVIDIA Canary 1B","kind":"Open model","benchmark":"Common Voice","language":"Spanish","condition":null,"year":null,"conversion":null,"claim":"NVIDIA Canary 1B had a word error rate of 3.99 percent on Common Voice Spanish.","quote":"WER on [MCV-16.1](https://commonvoice.mozilla.org/en/datasets) test set: | **Version** | **Model**     | **En**   | **De**   | **Es**   | **Fr**   | ... | 1.23.0  | canary-1b | 7.97 | 4.61 | 3.99 | 6.53 |","where_in_source":"WER table, es column","fact_url":"https://yapapp.cc/facts/nvidia-canary-1b-mcv-es-3-99","source_title":"canary-1b model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"n.d.","source_url":"https://huggingface.co/nvidia/canary-1b/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v3, FLEURS English","value":4,"low":null,"high":null,"as_written":"4","unit":"percent","system":"Whisper large-v3","kind":"Open model","benchmark":"FLEURS","language":"English","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, Whisper large-v3 had a word error rate of 4.00 percent on FLEURS English.","quote":"Model ar nl en fr de hi it pt es ... Whisper large-v3 15.44 5.87 4.00 5.55 5.46 28.87 2.71 3.90 2.81","where_in_source":"Table 4 (FLEURS), column en","fact_url":"https://yapapp.cc/facts/voxtral-2025-whisper-large-v3-fleurs-en-4-00","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v2, Italian","value":4,"low":null,"high":null,"as_written":"4","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Italian","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS read-speech benchmark, Whisper large-v2 had a word error rate of 4.0 percent in Italian.","quote":"Whisper large-v2 13.4 17.0 44.6 7.1 38.2 4.0 5.3","where_in_source":"Appendix D.2.4, FLEURS transcription table, Italian column","fact_url":"https://yapapp.cc/facts/whisper-fleurs-italian-4-0","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-17"},{"row":"Voxtral Small, FLEURS French","value":4.03,"low":null,"high":null,"as_written":"4.03","unit":"percent","system":"Voxtral Small","kind":"Open model","benchmark":"FLEURS","language":"French","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, Voxtral Small had a word error rate of 4.03 percent on FLEURS French.","quote":"Model ar nl en fr de hi it pt es ... Voxtral Small 13.44 4.94 3.35 4.03 3.38 7.69 2.62 3.79 2.72","where_in_source":"Table 4 (FLEURS), column fr","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-small-fleurs-fr-4-03","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini, LS test-other English","value":4.04,"low":null,"high":null,"as_written":"4.04","unit":"percent","system":"Voxtral Mini","kind":"Open model","benchmark":"LibriSpeech test-other","language":"English","condition":"Mixed benchmark","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, Voxtral Mini had a word error rate of 4.04 percent on LibriSpeech test-other.","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... Voxtral Mini 1.86 4.04 10.68 6.85 11.32 10.59 2.19 9.62 11.84","where_in_source":"Table 3 (short-form English), LibriSpeech test-other column","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-librispeech-test-other-4-04","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Canary-Qwen 2.5B, noise 5 dB","value":4.08,"low":null,"high":null,"as_written":"4.08","unit":"percent","system":"NVIDIA Canary Qwen 2.5B","kind":"Open model","benchmark":"LibriSpeech test-clean with white noise","language":"English","condition":"Noisy or distant","year":2025,"conversion":null,"claim":"NVIDIA Canary Qwen 2.5B had a word error rate of 4.08 percent on LibriSpeech test-clean with added white noise at 5 dB SNR.","quote":"WER on [Librispeech Test Clean](https://www.openslr.org/12) at different SNR (signal to noise ratio) levels of additive white noise | **Version** | **Model** | **SNR 10** | **SNR 5** | **SNR 0** | **SNR -5** | ... | 2.5.0       | Canary-Qwen-2.5B |    2.41%   |   4.08%   |   9.83%   |    30.60%  |","where_in_source":"White-noise table, SNR 5 dB column","fact_url":"https://yapapp.cc/facts/nvidia-canary-qwen-2-5b-white-noise-snr-5-db-4-08","source_title":"canary-qwen-2.5b model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/canary-qwen-2.5b/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Canary 1B Flash, CV German","value":4.09,"low":null,"high":null,"as_written":"4.09","unit":"percent","system":"NVIDIA Canary 1B Flash","kind":"Open model","benchmark":"Common Voice","language":"German","condition":null,"year":null,"conversion":null,"claim":"NVIDIA Canary 1B Flash had a word error rate of 4.09 percent on Common Voice German.","quote":"WER on [MCV-16.1](https://commonvoice.mozilla.org/en/datasets) test set: | **Version** | **Model**  |  **En**   | **De**   | **Es**   | **Fr**   | ... | nemo-main   | canary-1b-flash | 6.99 | 4.09 | 3.62 | 6.15 |","where_in_source":"WER table, de column","fact_url":"https://yapapp.cc/facts/nvidia-canary-1b-flash-mcv-de-4-09","source_title":"canary-1b-flash model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"n.d.","source_url":"https://huggingface.co/nvidia/canary-1b-flash/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Canary 1B, MLS French","value":4.12,"low":null,"high":null,"as_written":"4.12","unit":"percent","system":"NVIDIA Canary 1B","kind":"Open model","benchmark":"Multilingual LibriSpeech","language":"French","condition":"Clean read speech","year":null,"conversion":null,"claim":"NVIDIA Canary 1B had a word error rate of 4.12 percent on Multilingual LibriSpeech French.","quote":"WER on [MLS](https://huggingface.co/datasets/facebook/multilingual_librispeech) test set: | **Version** | **Model**     | **En**   | **De**   | **Es**   | **Fr**   | ... | 1.23.0  | canary-1b | 3.06 | 4.19 | 3.15 | 4.12 |","where_in_source":"WER table, fr column","fact_url":"https://yapapp.cc/facts/nvidia-canary-1b-mls-fr-4-12","source_title":"canary-1b model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"n.d.","source_url":"https://huggingface.co/nvidia/canary-1b/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini Transcribe, MLS French","value":4.14,"low":null,"high":null,"as_written":"4.14","unit":"percent","system":"Voxtral Mini Transcribe","kind":"Open model","benchmark":"Multilingual LibriSpeech","language":"French","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 6, Voxtral Mini Transcribe had a word error rate of 4.14 percent on Multilingual LibriSpeech French.","quote":"Model nl fr de it pt es ... Voxtral Mini Transcribe 9.63 4.14 5.64 9.28 5.17 3.87","where_in_source":"Table 6 (Multilingual LibriSpeech), column fr","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-transcribe-multilingual-librispeech-fr-4-14","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Canary 1B, MLS German","value":4.19,"low":null,"high":null,"as_written":"4.19","unit":"percent","system":"NVIDIA Canary 1B","kind":"Open model","benchmark":"Multilingual LibriSpeech","language":"German","condition":"Clean read speech","year":null,"conversion":null,"claim":"NVIDIA Canary 1B had a word error rate of 4.19 percent on Multilingual LibriSpeech German.","quote":"WER on [MLS](https://huggingface.co/datasets/facebook/multilingual_librispeech) test set: | **Version** | **Model**     | **En**   | **De**   | **Es**   | **Fr**   | ... | 1.23.0  | canary-1b | 3.06 | 4.19 | 3.15 | 4.12 |","where_in_source":"WER table, de column","fact_url":"https://yapapp.cc/facts/nvidia-canary-1b-mls-de-4-19","source_title":"canary-1b model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"n.d.","source_url":"https://huggingface.co/nvidia/canary-1b/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v2, English","value":4.2,"low":null,"high":null,"as_written":"4.2","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"English","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the Fleurs benchmark, the Whisper large-v2 model had a word error rate of 4.2 percent for English.","quote":"Model German Greek English Spanish Estonian Persian Finnish Tagalog French Galician Gujarati Hausa Hebrew Hindi ... Whisper large-v2 4.5 12.5 4.2 3.0 21.9 32.9 9.7 13.8 8.3 15.4 102.7 88.9 27.1 21.5","where_in_source":"Table 13 (Fleurs)","fact_url":"https://yapapp.cc/facts/whisper-fleurs-english-4-2","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-17"},{"row":"Voxtral Mini Transcribe, FLEURS French","value":4.22,"low":null,"high":null,"as_written":"4.22","unit":"percent","system":"Voxtral Mini Transcribe","kind":"Open model","benchmark":"FLEURS","language":"French","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, Voxtral Mini Transcribe had a word error rate of 4.22 percent on FLEURS French.","quote":"Model ar nl en fr de hi it pt es ... Voxtral Mini Transcribe 14.64 4.89 3.61 4.22 3.54 10.32 2.31 3.57 2.75","where_in_source":"Table 4 (FLEURS), column fr","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-transcribe-fleurs-fr-4-22","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Canary 1B v2, TED-LIUM English","value":4.29,"low":null,"high":null,"as_written":"4.29","unit":"percent","system":"NVIDIA Canary 1B v2","kind":"Open model","benchmark":"TED-LIUM","language":"English","condition":"Mixed benchmark","year":2025,"conversion":null,"claim":"NVIDIA Canary 1B v2 had a word error rate of 4.29 percent on the TED-LIUM test set in the Open ASR Leaderboard table.","quote":"|  **WER ↓**   | **RTFx**   | **Mean**   | **AMI**   | **GigaSpeech**   | **LS Clean**   | **LS Other**   | **Earnings22**   | **SPGISpech**   | **Tedlium**   | **Voxpopuli**   | ... | `Canary-1b-v2` | 749 | 7.15 | 16.01 | 10.82 | 2.18 | 3.56 | 11.79 | 2.28 | 4.29 | 6.25 |","where_in_source":"Open ASR Leaderboard table, TED-LIUM column","fact_url":"https://yapapp.cc/facts/nvidia-canary-1b-v2-ted-lium-open-asr-4-29","source_title":"canary-1b-v2 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/canary-1b-v2/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v2, Portuguese","value":4.3,"low":null,"high":null,"as_written":"4.3","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Portuguese","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS read-speech benchmark, Whisper large-v2 had a word error rate of 4.3 percent in Portuguese.","quote":"Whisper large-v2 6.7 75.3 102.4 5.4 93.7 4.3 14.4 5.6","where_in_source":"Appendix D.2.4, FLEURS transcription table, Portuguese column","fact_url":"https://yapapp.cc/facts/whisper-fleurs-portuguese-4-3","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-17"},{"row":"Canary 1B Flash, MLS German","value":4.36,"low":null,"high":null,"as_written":"4.36","unit":"percent","system":"NVIDIA Canary 1B Flash","kind":"Open model","benchmark":"Multilingual LibriSpeech","language":"German","condition":"Clean read speech","year":null,"conversion":null,"claim":"NVIDIA Canary 1B Flash had a word error rate of 4.36 percent on Multilingual LibriSpeech German.","quote":"WER on [MLS](https://huggingface.co/datasets/facebook/multilingual_librispeech) test set: | **Version** | **Model**  | **De**   | **Es**   | **Fr**   | ... | nemo-main   | canary-1b-flash | 4.36 | 2.69 | 4.47 |","where_in_source":"WER table, de column","fact_url":"https://yapapp.cc/facts/nvidia-canary-1b-flash-mls-de-4-36","source_title":"canary-1b-flash model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"n.d.","source_url":"https://huggingface.co/nvidia/canary-1b-flash/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini Transcribe, CV Italian","value":4.37,"low":null,"high":null,"as_written":"4.37","unit":"percent","system":"Voxtral Mini Transcribe","kind":"Open model","benchmark":"Common Voice","language":"Italian","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, Voxtral Mini Transcribe had a word error rate of 4.37 percent on Common Voice Italian.","quote":"Model ar nl en fr de hi it pt es ... Voxtral Mini Transcribe 62.01 4.71 8.25 7.29 4.85 10.42 4.37 6.70 3.96","where_in_source":"Table 5 (Common Voice), column it","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-transcribe-common-voice-it-4-37","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, MLS Spanish","value":4.39,"low":null,"high":null,"as_written":"4.39","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"Multilingual LibriSpeech","language":"Spanish","condition":"Clean read speech","year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 4.39 percent on the Multilingual LibriSpeech Spanish test.","quote":"| Language | Fleurs | MLS | CoVoST | ... | **es** | 3.45% | 4.39% | 3.41% |","where_in_source":"Multilingual ASR table, MLS column, es row","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-multilingual-librispeech-es-4-39","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini, FLEURS German","value":4.4,"low":null,"high":null,"as_written":"4.4","unit":"percent","system":"Voxtral Mini","kind":"Open model","benchmark":"FLEURS","language":"German","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, Voxtral Mini had a word error rate of 4.40 percent on FLEURS German.","quote":"Model ar nl en fr de hi it pt es ... Voxtral Mini 25.40 6.27 3.77 4.87 4.40 9.26 2.51 3.76 3.52","where_in_source":"Table 4 (FLEURS), column de","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-fleurs-de-4-40","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Canary 1B Flash, MLS French","value":4.47,"low":null,"high":null,"as_written":"4.47","unit":"percent","system":"NVIDIA Canary 1B Flash","kind":"Open model","benchmark":"Multilingual LibriSpeech","language":"French","condition":"Clean read speech","year":null,"conversion":null,"claim":"NVIDIA Canary 1B Flash had a word error rate of 4.47 percent on Multilingual LibriSpeech French.","quote":"WER on [MLS](https://huggingface.co/datasets/facebook/multilingual_librispeech) test set: | **Version** | **Model**  | **De**   | **Es**   | **Fr**   | ... | nemo-main   | canary-1b-flash | 4.36 | 2.69 | 4.47 |","where_in_source":"WER table, fr column","fact_url":"https://yapapp.cc/facts/nvidia-canary-1b-flash-mls-fr-4-47","source_title":"canary-1b-flash model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"n.d.","source_url":"https://huggingface.co/nvidia/canary-1b-flash/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, English (our test)","value":4.48,"low":null,"high":null,"as_written":"4.48","unit":"percent","system":"Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"FLEURS","language":"English","condition":"Clean read speech","year":2026,"conversion":null,"claim":"In our October 2026 benchmark on a Mac with an M2 Pro, Parakeet TDT 0.6B v3 had a word error rate of 4.48 percent on 100 English recordings from the FLEURS test set.","quote":null,"where_in_source":null,"fact_url":"https://yapapp.cc/facts/m-fleurs-2026-10-parakeet-en-wer","source_title":null,"source_publisher":null,"source_year":null,"source_url":null,"source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v2, German","value":4.5,"low":null,"high":null,"as_written":"4.5","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"German","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the Fleurs benchmark, the Whisper large-v2 model had a word error rate of 4.5 percent for German.","quote":"Model German Greek English Spanish Estonian Persian Finnish Tagalog French Galician Gujarati Hausa Hebrew Hindi ... Whisper large-v2 4.5 12.5 4.2 3.0 21.9 32.9 9.7 13.8 8.3 15.4 102.7 88.9 27.1 21.5","where_in_source":"Table 13 (Fleurs)","fact_url":"https://yapapp.cc/facts/whisper-fleurs-german-4-5","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-16"},{"row":"Whisper large-v3-turbo, Russian (our test)","value":4.59,"low":null,"high":null,"as_written":"4.59","unit":"percent","system":"Whisper large-v3-turbo","kind":"Open model","benchmark":"FLEURS","language":"Russian","condition":"Clean read speech","year":2026,"conversion":null,"claim":"In our October 2026 benchmark on a Mac with an M2 Pro, Whisper large-v3-turbo had a word error rate of 4.59 percent on 100 Russian recordings from the FLEURS test set.","quote":null,"where_in_source":null,"fact_url":"https://yapapp.cc/facts/m-fleurs-2026-10-whisper-ru-wer","source_title":null,"source_publisher":null,"source_year":null,"source_url":null,"source_doi":null,"checked":"2026-10-19"},{"row":"Canary 1B, CV German","value":4.61,"low":null,"high":null,"as_written":"4.61","unit":"percent","system":"NVIDIA Canary 1B","kind":"Open model","benchmark":"Common Voice","language":"German","condition":null,"year":null,"conversion":null,"claim":"NVIDIA Canary 1B had a word error rate of 4.61 percent on Common Voice German.","quote":"WER on [MCV-16.1](https://commonvoice.mozilla.org/en/datasets) test set: | **Version** | **Model**     | **En**   | **De**   | **Es**   | **Fr**   | ... | 1.23.0  | canary-1b | 7.97 | 4.61 | 3.99 | 6.53 |","where_in_source":"WER table, de column","fact_url":"https://yapapp.cc/facts/nvidia-canary-1b-mcv-de-4-61","source_title":"canary-1b model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"n.d.","source_url":"https://huggingface.co/nvidia/canary-1b/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini Transcribe, CV Dutch","value":4.71,"low":null,"high":null,"as_written":"4.71","unit":"percent","system":"Voxtral Mini Transcribe","kind":"Open model","benchmark":"Common Voice","language":"Dutch","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, Voxtral Mini Transcribe had a word error rate of 4.71 percent on Common Voice Dutch.","quote":"Model ar nl en fr de hi it pt es ... Voxtral Mini Transcribe 62.01 4.71 8.25 7.29 4.85 10.42 4.37 6.70 3.96","where_in_source":"Table 5 (Common Voice), column nl","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-transcribe-common-voice-nl-4-71","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, Russian (our test)","value":4.74,"low":null,"high":null,"as_written":"4.74","unit":"percent","system":"Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"FLEURS","language":"Russian","condition":"Clean read speech","year":2026,"conversion":null,"claim":"In our October 2026 benchmark on a Mac with an M2 Pro, Parakeet TDT 0.6B v3 had a word error rate of 4.74 percent on 100 Russian recordings from the FLEURS test set.","quote":null,"where_in_source":null,"fact_url":"https://yapapp.cc/facts/m-fleurs-2026-10-parakeet-ru-wer","source_title":null,"source_publisher":null,"source_year":null,"source_url":null,"source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, FLEURS Portuguese","value":4.76,"low":null,"high":null,"as_written":"4.76","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"FLEURS","language":"Portuguese","condition":"Clean read speech","year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 4.76 percent on the FLEURS Portuguese test.","quote":"| Language | Fleurs | MLS | CoVoST | ... | **pt** | 4.76% | 7.50% | 3.96% |","where_in_source":"Multilingual ASR table, Fleurs column, pt row","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-fleurs-pt-4-76","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"wav2vec 2.0, 10 min labeled","value":4.8,"low":null,"high":null,"as_written":"4.8","unit":"percent","system":"wav2vec 2.0 with ten minutes of labeled data","kind":"Open model","benchmark":"LibriSpeech test-clean","language":"English","condition":"Clean read speech","year":2020,"conversion":"The fact also gives 8.2 on test-other; that figure is a separate row. Language is taken from the test set name: LibriSpeech is English.","claim":"With just ten minutes of labeled data and pre-training on 53,000 hours of unlabeled audio, wav2vec 2.0 still reached 4.8 percent word error rate on LibriSpeech clean and 8.2 percent on other.","quote":"Using just ten minutes of labeled data and pre-training on 53k hours of unlabeled data still achieves 4.8/8.2 WER.","where_in_source":"Abstract","fact_url":"https://yapapp.cc/facts/wav2vec2-ten-minutes-labeled-4-8-8-2","source_title":"wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations","source_publisher":"arXiv (NeurIPS 2020)","source_year":"2020","source_url":"https://arxiv.org/abs/2006.11477","source_doi":null,"checked":"2026-10-16"},{"row":"Parakeet TDT v3, German (our test)","value":4.84,"low":null,"high":null,"as_written":"4.84","unit":"percent","system":"Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"FLEURS","language":"German","condition":"Clean read speech","year":2026,"conversion":null,"claim":"In our October 2026 benchmark on a Mac with an M2 Pro, Parakeet TDT 0.6B v3 had a word error rate of 4.84 percent on 100 German recordings from the FLEURS test set.","quote":null,"where_in_source":null,"fact_url":"https://yapapp.cc/facts/m-fleurs-2026-10-parakeet-de-wer","source_title":null,"source_publisher":null,"source_year":null,"source_url":null,"source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, CoVoST 2 German","value":4.84,"low":null,"high":null,"as_written":"4.84","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"CoVoST 2","language":"German","condition":null,"year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 4.84 percent on the CoVoST 2 German test.","quote":"| Language | Fleurs | MLS | CoVoST | ... | **de** | 5.04% | - | 4.84% |","where_in_source":"Multilingual ASR table, CoVoST column, de row","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-covost-2-de-4-84","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, FLEURS English","value":4.85,"low":null,"high":null,"as_written":"4.85","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"FLEURS","language":"English","condition":"Clean read speech","year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 4.85 percent on the FLEURS English test.","quote":"| Language | Fleurs | MLS | CoVoST | ... | **en** | 4.85% | - | 6.80% |","where_in_source":"Multilingual ASR table, Fleurs column, en row","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-fleurs-en-4-85","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini Transcribe, CV German","value":4.85,"low":null,"high":null,"as_written":"4.85","unit":"percent","system":"Voxtral Mini Transcribe","kind":"Open model","benchmark":"Common Voice","language":"German","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, Voxtral Mini Transcribe had a word error rate of 4.85 percent on Common Voice German.","quote":"Model ar nl en fr de hi it pt es ... Voxtral Mini Transcribe 62.01 4.71 8.25 7.29 4.85 10.42 4.37 6.70 3.96","where_in_source":"Table 5 (Common Voice), column de","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-transcribe-common-voice-de-4-85","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini, FLEURS French","value":4.87,"low":null,"high":null,"as_written":"4.87","unit":"percent","system":"Voxtral Mini","kind":"Open model","benchmark":"FLEURS","language":"French","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, Voxtral Mini had a word error rate of 4.87 percent on FLEURS French.","quote":"Model ar nl en fr de hi it pt es ... Voxtral Mini 25.40 6.27 3.77 4.87 4.40 9.26 2.51 3.76 3.52","where_in_source":"Table 4 (FLEURS), column fr","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-fleurs-fr-4-87","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini Transcribe, FLEURS Dutch","value":4.89,"low":null,"high":null,"as_written":"4.89","unit":"percent","system":"Voxtral Mini Transcribe","kind":"Open model","benchmark":"FLEURS","language":"Dutch","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, Voxtral Mini Transcribe had a word error rate of 4.89 percent on FLEURS Dutch.","quote":"Model ar nl en fr de hi it pt es ... Voxtral Mini Transcribe 14.64 4.89 3.61 4.22 3.54 10.32 2.31 3.57 2.75","where_in_source":"Table 4 (FLEURS), column nl","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-transcribe-fleurs-nl-4-89","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Small, FLEURS Dutch","value":4.94,"low":null,"high":null,"as_written":"4.94","unit":"percent","system":"Voxtral Small","kind":"Open model","benchmark":"FLEURS","language":"Dutch","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, Voxtral Small had a word error rate of 4.94 percent on FLEURS Dutch.","quote":"Model ar nl en fr de hi it pt es ... Voxtral Small 13.44 4.94 3.35 4.03 3.38 7.69 2.62 3.79 2.72","where_in_source":"Table 4 (FLEURS), column nl","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-small-fleurs-nl-4-94","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, MLS French","value":4.97,"low":null,"high":null,"as_written":"4.97","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"Multilingual LibriSpeech","language":"French","condition":"Clean read speech","year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 4.97 percent on the Multilingual LibriSpeech French test.","quote":"| Language | Fleurs | MLS | CoVoST | ... | **fr** | 5.15% | 4.97% | 6.05% |","where_in_source":"Multilingual ASR table, MLS column, fr row","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-multilingual-librispeech-fr-4-97","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini, CV Spanish","value":4.98,"low":null,"high":null,"as_written":"4.98","unit":"percent","system":"Voxtral Mini","kind":"Open model","benchmark":"Common Voice","language":"Spanish","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, Voxtral Mini had a word error rate of 4.98 percent on Common Voice Spanish.","quote":"Model ar nl en fr de hi it pt es ... Voxtral Mini 63.98 6.03 10.22 8.92 6.07 12.74 5.91 7.76 4.98","where_in_source":"Table 5 (Common Voice), column es","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-common-voice-es-4-98","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"microsoft/Phi-4-multimodal-instruct","value":5.02125,"low":null,"high":null,"as_written":"5.02","unit":"percent","system":"microsoft/Phi-4-multimodal-instruct","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: mit): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, microsoft/Phi-4-multimodal-instruct had an average word error rate of 5.02 percent.","quote":"microsoft/Phi-4-multimodal-instruct,5.02125,162.53","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-microsoft-phi-4-multimodal-instruct-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, FLEURS German","value":5.04,"low":null,"high":null,"as_written":"5.04","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"FLEURS","language":"German","condition":"Clean read speech","year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 5.04 percent on the FLEURS German test.","quote":"| Language | Fleurs | MLS | CoVoST | ... | **de** | 5.04% | - | 4.84% |","where_in_source":"Multilingual ASR table, Fleurs column, de row","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-fleurs-de-5-04","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v3-turbo, German (our test)","value":5.08,"low":null,"high":null,"as_written":"5.08","unit":"percent","system":"Whisper large-v3-turbo","kind":"Open model","benchmark":"FLEURS","language":"German","condition":"Clean read speech","year":2026,"conversion":null,"claim":"In our October 2026 benchmark on a Mac with an M2 Pro, Whisper large-v3-turbo had a word error rate of 5.08 percent on 100 German recordings from the FLEURS test set.","quote":null,"where_in_source":null,"fact_url":"https://yapapp.cc/facts/m-fleurs-2026-10-whisper-de-wer","source_title":null,"source_publisher":null,"source_year":null,"source_url":null,"source_doi":null,"checked":"2026-10-19"},{"row":"Canary 1B v2, noise 0 dB","value":5.08,"low":null,"high":null,"as_written":"5.08","unit":"percent","system":"NVIDIA Canary 1B v2","kind":"Open model","benchmark":"LibriSpeech test-clean with MUSAN noise","language":"English","condition":"Noisy or distant","year":2025,"conversion":null,"claim":"NVIDIA Canary 1B v2 had a word error rate of 5.08 percent on LibriSpeech Clean with MUSAN noise at 0 dB SNR.","quote":"Performance across different Signal-to-Noise Ratios (SNR) using MUSAN music and noise samples \\[16] on the [LibriSpeech Clean test set](https://www.openslr.org/12). ... | **SNR (dB)**       | 100      | 10    | 5     | 0     | -5     | ... | **`Canary-1b-v2`** | 2.18% | 2.29% | 2.80% | 5.08% | 19.38% |","where_in_source":"Noise robustness table, SNR 0 dB column","fact_url":"https://yapapp.cc/facts/nvidia-canary-1b-v2-musan-snr-0-db-5-08","source_title":"canary-1b-v2 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/canary-1b-v2/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v3, MLS French","value":5.09,"low":null,"high":null,"as_written":"5.09","unit":"percent","system":"Whisper large-v3","kind":"Open model","benchmark":"Multilingual LibriSpeech","language":"French","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 6, Whisper large-v3 had a word error rate of 5.09 percent on Multilingual LibriSpeech French.","quote":"Model nl fr de it pt es ... Whisper large-v3 9.19 5.09 5.72 9.78 7.03 3.89","where_in_source":"Table 6 (Multilingual LibriSpeech), column fr","fact_url":"https://yapapp.cc/facts/voxtral-2025-whisper-large-v3-multilingual-librispeech-fr-5-09","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, CoVoST 2 Ukrainian","value":5.1,"low":null,"high":null,"as_written":"5.1","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"CoVoST 2","language":"Ukrainian","condition":null,"year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 5.10 percent on the CoVoST 2 Ukrainian test.","quote":"| Language | Fleurs | MLS | CoVoST | ... | **uk** | 6.79% | - | 5.10% |","where_in_source":"Multilingual ASR table, CoVoST column, uk row","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-covost-2-uk-5-10","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini, MLS Spanish","value":5.12,"low":null,"high":null,"as_written":"5.12","unit":"percent","system":"Voxtral Mini","kind":"Open model","benchmark":"Multilingual LibriSpeech","language":"Spanish","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 6, Voxtral Mini had a word error rate of 5.12 percent on Multilingual LibriSpeech Spanish.","quote":"Model nl fr de it pt es ... Voxtral Mini 10.09 5.28 7.09 11.30 6.72 5.12","where_in_source":"Table 6 (Multilingual LibriSpeech), column es","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-multilingual-librispeech-es-5-12","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, FLEURS French","value":5.15,"low":null,"high":null,"as_written":"5.15","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"FLEURS","language":"French","condition":"Clean read speech","year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 5.15 percent on the FLEURS French test.","quote":"| Language | Fleurs | MLS | CoVoST | ... | **fr** | 5.15% | 4.97% | 6.05% |","where_in_source":"Multilingual ASR table, Fleurs column, fr row","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-fleurs-fr-5-15","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini Transcribe, MLS Portuguese","value":5.17,"low":null,"high":null,"as_written":"5.17","unit":"percent","system":"Voxtral Mini Transcribe","kind":"Open model","benchmark":"Multilingual LibriSpeech","language":"Portuguese","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 6, Voxtral Mini Transcribe had a word error rate of 5.17 percent on Multilingual LibriSpeech Portuguese.","quote":"Model nl fr de it pt es ... Voxtral Mini Transcribe 9.63 4.14 5.64 9.28 5.17 3.87","where_in_source":"Table 6 (Multilingual LibriSpeech), column pt","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-transcribe-multilingual-librispeech-pt-5-17","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v2, LibriSpeech other","value":5.2,"low":null,"high":null,"as_written":"5.2","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"LibriSpeech test-other","language":"English","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the LibriSpeech test-other split, the Whisper large-v2 model had a word error rate of 5.2 percent, in the paper's English table with greedy decoding.","quote":"Whisper large-v2 2.7 5.2 4.0 3.9 17.6 13.8 9.0 6.2 16.2 25.5 16.9 36.4 7.3 4.4","where_in_source":"Table D.1.1 (English transcription, greedy decoding), Whisper large-v2 row, column LibriSpeech test-other","fact_url":"https://yapapp.cc/facts/whisper-librispeech-other-wer-5-2","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-19"},{"row":"ibm-granite/granite-speech-3.3-8b","value":5.2625,"low":null,"high":null,"as_written":"5.26","unit":"percent","system":"ibm-granite/granite-speech-3.3-8b","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: apache-2.0): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, ibm-granite/granite-speech-3.3-8b had an average word error rate of 5.26 percent.","quote":"ibm-granite/granite-speech-3.3-8b,5.2625,262.79","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-ibm-granite-granite-speech-3-3-8b-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini, MLS French","value":5.28,"low":null,"high":null,"as_written":"5.28","unit":"percent","system":"Voxtral Mini","kind":"Open model","benchmark":"Multilingual LibriSpeech","language":"French","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 6, Voxtral Mini had a word error rate of 5.28 percent on Multilingual LibriSpeech French.","quote":"Model nl fr de it pt es ... Voxtral Mini 10.09 5.28 7.09 11.30 6.72 5.12","where_in_source":"Table 6 (Multilingual LibriSpeech), column fr","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-multilingual-librispeech-fr-5-28","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, French (our test)","value":5.29,"low":null,"high":null,"as_written":"5.29","unit":"percent","system":"Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"FLEURS","language":"French","condition":"Clean read speech","year":2026,"conversion":null,"claim":"In our October 2026 benchmark on a Mac with an M2 Pro, Parakeet TDT 0.6B v3 had a word error rate of 5.29 percent on 100 French recordings from the FLEURS test set.","quote":null,"where_in_source":null,"fact_url":"https://yapapp.cc/facts/m-fleurs-2026-10-parakeet-fr-wer","source_title":null,"source_publisher":null,"source_year":null,"source_url":null,"source_doi":null,"checked":"2026-10-19"},{"row":"IBM Granite Speech 4.1 2B","value":5.33,"low":null,"high":null,"as_written":"5.33","unit":"percent","system":"IBM Granite Speech 4.1 2B","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":null,"claim":"In a Benchmarklist mirror of the Open ASR Leaderboard, dated May 6, 2026, IBM Granite Speech 4.1 2B had the lowest average word error rate among open models, at 5.33 percent, 0.09 points ahead of Cohere Transcribe 03-2026.","quote":"IBM Granite Speech 4.1 2B leads with 5.33% WER, edging out Cohere Transcribe 03-2026 by 0.09 percentage points","where_in_source":"Results summary, 'What the Results Say'","fact_url":"https://yapapp.cc/facts/benchmarklist-2026-granite-speech-41-5-33-wer","source_title":"Open ASR Leaderboard (Benchmarklist mirror, results updated May 6, 2026)","source_publisher":"Benchmarklist","source_year":"2026","source_url":"https://benchmarklist.com/benchmarks/open_asr_leaderboard/","source_doi":null,"checked":"2026-10-17"},{"row":"ibm-granite/granite-speech-3.3-2b","value":5.3575,"low":null,"high":null,"as_written":"5.36","unit":"percent","system":"ibm-granite/granite-speech-3.3-2b","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: apache-2.0): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, ibm-granite/granite-speech-3.3-2b had an average word error rate of 5.36 percent.","quote":"ibm-granite/granite-speech-3.3-2b,5.3575,509.35","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-ibm-granite-granite-speech-3-3-2b-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v2, Polish","value":5.4,"low":null,"high":null,"as_written":"5.4","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Polish","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 5.4 percent for Polish.","quote":"Whisper large-v2 6.7 75.3 102.4 5.4 93.7 4.3 14.4 5.6 156.5 11.7 23.1 121.0 102.9 33.9","where_in_source":"Table 13 (FLEURS), Whisper large-v2 row, column Polish","fact_url":"https://yapapp.cc/facts/whisper-fleurs-polish-5-4","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-19"},{"row":"Cohere Labs Transcribe","value":5.42,"low":null,"high":null,"as_written":"5.42","unit":"percent","system":"Cohere Labs Transcribe","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2025,"conversion":null,"claim":"On the Open ASR Leaderboard's English short-form tests, Cohere Labs Transcribe had an average word error rate of 5.42 percent, and the paper's table marks it as open.","quote":"Cohere Labs Transcribe Yes 5.42 525 FastConformer","where_in_source":"Table 3, short-form English","fact_url":"https://yapapp.cc/facts/open-asr-cohere-transcribe-5-42","source_title":"Open ASR Leaderboard: Towards Reproducible and Transparent Multilingual and Long-Form Speech Recognition Evaluation","source_publisher":"arXiv","source_year":"2025","source_url":"https://arxiv.org/html/2510.06961v4","source_doi":null,"checked":"2026-10-17"},{"row":"Whisper large-v3, FLEURS German","value":5.46,"low":null,"high":null,"as_written":"5.46","unit":"percent","system":"Whisper large-v3","kind":"Open model","benchmark":"FLEURS","language":"German","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, Whisper large-v3 had a word error rate of 5.46 percent on FLEURS German.","quote":"Model ar nl en fr de hi it pt es ... Whisper large-v3 15.44 5.87 4.00 5.55 5.46 28.87 2.71 3.90 2.81","where_in_source":"Table 4 (FLEURS), column de","fact_url":"https://yapapp.cc/facts/voxtral-2025-whisper-large-v3-fleurs-de-5-46","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, FLEURS Russian","value":5.51,"low":null,"high":null,"as_written":"5.51","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"FLEURS","language":"Russian","condition":"Clean read speech","year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 5.51 percent on the FLEURS Russian test.","quote":"| Language | Fleurs | MLS | CoVoST | ... | **ru** | 5.51% | - | 3.00% |","where_in_source":"Multilingual ASR table, Fleurs column, ru row","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-fleurs-ru-5-51","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v3, FLEURS French","value":5.55,"low":null,"high":null,"as_written":"5.55","unit":"percent","system":"Whisper large-v3","kind":"Open model","benchmark":"FLEURS","language":"French","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, Whisper large-v3 had a word error rate of 5.55 percent on FLEURS French.","quote":"Model ar nl en fr de hi it pt es ... Whisper large-v3 15.44 5.87 4.00 5.55 5.46 28.87 2.71 3.90 2.81","where_in_source":"Table 4 (FLEURS), column fr","fact_url":"https://yapapp.cc/facts/voxtral-2025-whisper-large-v3-fleurs-fr-5-55","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Small, MLS German","value":5.57,"low":null,"high":null,"as_written":"5.57","unit":"percent","system":"Voxtral Small","kind":"Open model","benchmark":"Multilingual LibriSpeech","language":"German","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 6, Voxtral Small had a word error rate of 5.57 percent on Multilingual LibriSpeech German.","quote":"Model nl fr de it pt es ... Voxtral Small 9.43 3.73 5.57 8.44 5.85 3.62","where_in_source":"Table 6 (Multilingual LibriSpeech), column de","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-small-multilingual-librispeech-de-5-57","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v2, Russian","value":5.6,"low":null,"high":null,"as_written":"5.6","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Russian","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the Fleurs benchmark, the Whisper large-v2 model had a word error rate of 5.6 percent for Russian.","quote":"Model Dutch Occitan Punjabi Polish Pashto Portuguese Romanian Russian Sindhi Slovak Slovenian Shona Somali Serbian ... Whisper large-v2 6.7 75.3 102.4 5.4 93.7 4.3 14.4 5.6 156.5 11.7 23.1 121.0 102.9 33.9","where_in_source":"Table 13 (Fleurs)","fact_url":"https://yapapp.cc/facts/whisper-fleurs-russian-5-6","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-16"},{"row":"Canary 1B Flash, VoxPopuli English","value":5.63,"low":null,"high":null,"as_written":"5.63","unit":"percent","system":"NVIDIA Canary 1B Flash","kind":"Open model","benchmark":"VoxPopuli","language":"English","condition":"Mixed benchmark","year":null,"conversion":null,"claim":"NVIDIA Canary 1B Flash had a word error rate of 5.63 percent on the VoxPopuli test set in the Open ASR Leaderboard table.","quote":"| **Version** | **Model**     | **RTFx**   | **AMI**   | **GigaSpeech**   | **LS Clean**   | **LS Other**   | **Earnings22**   | **SPGISpech**   | **Tedlium**   | **Voxpopuli**   | ... | nemo-main  | canary-1b-flash | 1045.75 | 13.11 | 9.85 | 1.48 | 2.87 | 12.79 | 1.95 | 3.12 | 5.63 |","where_in_source":"Open ASR Leaderboard table, VoxPopuli column","fact_url":"https://yapapp.cc/facts/nvidia-canary-1b-flash-voxpopuli-open-asr-5-63","source_title":"canary-1b-flash model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"n.d.","source_url":"https://huggingface.co/nvidia/canary-1b-flash/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Canary-Qwen 2.5B, Open ASR avg","value":5.63,"low":null,"high":null,"as_written":"5.63","unit":"percent","system":"NVIDIA Canary Qwen 2.5B","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2025,"conversion":null,"claim":"On the same leaderboard track, NVIDIA Canary Qwen 2.5B, a speech encoder with a language-model decoder, had an average word error rate of 5.63 percent and an RTFx of 418.","quote":"NVIDIA Canary Qwen 2.5B Yes 5.63 418 FastConformer","where_in_source":"Table 3","fact_url":"https://yapapp.cc/facts/open-asr-canary-qwen-5-63-rtfx-418","source_title":"Open ASR Leaderboard: Towards Reproducible and Transparent Multilingual and Long-Form Speech Recognition Evaluation","source_publisher":"arXiv","source_year":"2025","source_url":"https://arxiv.org/html/2510.06961v4","source_doi":null,"checked":"2026-10-16"},{"row":"Voxtral Mini Transcribe, MLS German","value":5.64,"low":null,"high":null,"as_written":"5.64","unit":"percent","system":"Voxtral Mini Transcribe","kind":"Open model","benchmark":"Multilingual LibriSpeech","language":"German","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 6, Voxtral Mini Transcribe had a word error rate of 5.64 percent on Multilingual LibriSpeech German.","quote":"Model nl fr de it pt es ... Voxtral Mini Transcribe 9.63 4.14 5.64 9.28 5.17 3.87","where_in_source":"Table 6 (Multilingual LibriSpeech), column de","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-transcribe-multilingual-librispeech-de-5-64","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, Ukrainian (our test)","value":5.65,"low":null,"high":null,"as_written":"5.65","unit":"percent","system":"Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"FLEURS","language":"Ukrainian","condition":"Clean read speech","year":2026,"conversion":null,"claim":"In our October 2026 benchmark on a Mac with an M2 Pro, Parakeet TDT 0.6B v3 had a word error rate of 5.65 percent on 100 Ukrainian recordings from the FLEURS test set.","quote":null,"where_in_source":null,"fact_url":"https://yapapp.cc/facts/m-fleurs-2026-10-parakeet-uk-wer","source_title":null,"source_publisher":null,"source_year":null,"source_url":null,"source_doi":null,"checked":"2026-10-19"},{"row":"Canary-Qwen 2.5B, VoxPopuli English","value":5.66,"low":null,"high":null,"as_written":"5.66","unit":"percent","system":"NVIDIA Canary Qwen 2.5B","kind":"Open model","benchmark":"VoxPopuli","language":"English","condition":"Mixed benchmark","year":2025,"conversion":null,"claim":"NVIDIA Canary Qwen 2.5B had a word error rate of 5.66 percent on the VoxPopuli test set in the Open ASR Leaderboard table.","quote":"| **Version** | **Model**     | **RTFx**   | **Mean**   | **AMI**   | **GigaSpeech**   | **LS Clean**   | **LS Other**   | **Earnings22**   | **SPGISpech**   | **Tedlium**   | **Voxpopuli**   | ... | 2.5.0  | Canary-Qwen-2.5B | 418 | 5.63 | 10.18 | 9.41 | 1.60 | 3.10 | 10.42 | 1.90 | 2.72 | 5.66 |","where_in_source":"Open ASR Leaderboard table, VoxPopuli column","fact_url":"https://yapapp.cc/facts/nvidia-canary-qwen-2-5b-voxpopuli-open-asr-5-66","source_title":"canary-qwen-2.5b model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/canary-qwen-2.5b/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v3, CV Spanish","value":5.66,"low":null,"high":null,"as_written":"5.66","unit":"percent","system":"Whisper large-v3","kind":"Open model","benchmark":"Common Voice","language":"Spanish","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, Whisper large-v3 had a word error rate of 5.66 percent on Common Voice Spanish.","quote":"Model ar nl en fr de hi it pt es ... Whisper large-v3 50.58 5.83 22.91 11.33 6.25 46.75 6.81 7.17 5.66","where_in_source":"Table 5 (Common Voice), column es","fact_url":"https://yapapp.cc/facts/voxtral-2025-whisper-large-v3-common-voice-es-5-66","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Vosk en-us-0.22, 1.8 GB","value":5.69,"low":null,"high":null,"as_written":"5.69","unit":"percent","system":"Vosk vosk-model-en-us-0.22","kind":"Open model","benchmark":"LibriSpeech test-clean","language":"English","condition":"Clean read speech","year":2026,"conversion":null,"claim":"Vosk's 1.8 GB English model vosk-model-en-us-0.22 had a word error rate of 5.69 percent on LibriSpeech test-clean, and it is listed under the Apache 2.0 license.","quote":"vosk-model-en-us-0.22 1.8G 5.69 (librispeech test-clean) 6.05 (tedlium) 29.78(callcenter) Accurate generic US English model Apache 2.0","where_in_source":"Model list table","fact_url":"https://yapapp.cc/facts/vosk-en-us-0-22-wer-5-69","source_title":"Vosk models (Alpha Cephei)","source_publisher":"Alpha Cephei","source_year":"2026","source_url":"https://alphacephei.com/vosk/models","source_doi":null,"checked":"2026-10-17"},{"row":"Whisper large-v3, MLS German","value":5.72,"low":null,"high":null,"as_written":"5.72","unit":"percent","system":"Whisper large-v3","kind":"Open model","benchmark":"Multilingual LibriSpeech","language":"German","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 6, Whisper large-v3 had a word error rate of 5.72 percent on Multilingual LibriSpeech German.","quote":"Model nl fr de it pt es ... Whisper large-v3 9.19 5.09 5.72 9.78 7.03 3.89","where_in_source":"Table 6 (Multilingual LibriSpeech), column de","fact_url":"https://yapapp.cc/facts/voxtral-2025-whisper-large-v3-multilingual-librispeech-de-5-72","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Qwen/Qwen3-ASR-1.7B-hf","value":5.745714285714286,"low":null,"high":null,"as_written":"5.75","unit":"percent","system":"Qwen/Qwen3-ASR-1.7B-hf","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: apache-2.0): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, Qwen/Qwen3-ASR-1.7B-hf had an average word error rate of 5.75 percent.","quote":"Qwen/Qwen3-ASR-1.7B-hf,5.745714285714286,835.62","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-qwen-qwen3-asr-1-7b-hf-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v3, CV Dutch","value":5.83,"low":null,"high":null,"as_written":"5.83","unit":"percent","system":"Whisper large-v3","kind":"Open model","benchmark":"Common Voice","language":"Dutch","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, Whisper large-v3 had a word error rate of 5.83 percent on Common Voice Dutch.","quote":"Model ar nl en fr de hi it pt es ... Whisper large-v3 50.58 5.83 22.91 11.33 6.25 46.75 6.81 7.17 5.66","where_in_source":"Table 5 (Common Voice), column nl","fact_url":"https://yapapp.cc/facts/voxtral-2025-whisper-large-v3-common-voice-nl-5-83","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Small, MLS Portuguese","value":5.85,"low":null,"high":null,"as_written":"5.85","unit":"percent","system":"Voxtral Small","kind":"Open model","benchmark":"Multilingual LibriSpeech","language":"Portuguese","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 6, Voxtral Small had a word error rate of 5.85 percent on Multilingual LibriSpeech Portuguese.","quote":"Model nl fr de it pt es ... Voxtral Small 9.43 3.73 5.57 8.44 5.85 3.62","where_in_source":"Table 6 (Multilingual LibriSpeech), column pt","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-small-multilingual-librispeech-pt-5-85","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v3, FLEURS Dutch","value":5.87,"low":null,"high":null,"as_written":"5.87","unit":"percent","system":"Whisper large-v3","kind":"Open model","benchmark":"FLEURS","language":"Dutch","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, Whisper large-v3 had a word error rate of 5.87 percent on FLEURS Dutch.","quote":"Model ar nl en fr de hi it pt es ... Whisper large-v3 15.44 5.87 4.00 5.55 5.46 28.87 2.71 3.90 2.81","where_in_source":"Table 4 (FLEURS), column nl","fact_url":"https://yapapp.cc/facts/voxtral-2025-whisper-large-v3-fleurs-nl-5-87","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini, CV Italian","value":5.91,"low":null,"high":null,"as_written":"5.91","unit":"percent","system":"Voxtral Mini","kind":"Open model","benchmark":"Common Voice","language":"Italian","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, Voxtral Mini had a word error rate of 5.91 percent on Common Voice Italian.","quote":"Model ar nl en fr de hi it pt es ... Voxtral Mini 63.98 6.03 10.22 8.92 6.07 12.74 5.91 7.76 4.98","where_in_source":"Table 5 (Common Voice), column it","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-common-voice-it-5-91","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini, CV Dutch","value":6.03,"low":null,"high":null,"as_written":"6.03","unit":"percent","system":"Voxtral Mini","kind":"Open model","benchmark":"Common Voice","language":"Dutch","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, Voxtral Mini had a word error rate of 6.03 percent on Common Voice Dutch.","quote":"Model ar nl en fr de hi it pt es ... Voxtral Mini 63.98 6.03 10.22 8.92 6.07 12.74 5.91 7.76 4.98","where_in_source":"Table 5 (Common Voice), column nl","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-common-voice-nl-6-03","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, CoVoST 2 French","value":6.05,"low":null,"high":null,"as_written":"6.05","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"CoVoST 2","language":"French","condition":null,"year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 6.05 percent on the CoVoST 2 French test.","quote":"| Language | Fleurs | MLS | CoVoST | ... | **fr** | 5.15% | 4.97% | 6.05% |","where_in_source":"Multilingual ASR table, CoVoST column, fr row","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-covost-2-fr-6-05","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"NVIDIA Parakeet TDT 0.6B v2","value":6.05,"low":null,"high":null,"as_written":"6.05","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v2","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2025,"conversion":null,"claim":"On the Open ASR Leaderboard's English short-form track, NVIDIA Parakeet TDT 0.6B v2 had an average word error rate of 6.05 percent and an inverse real-time factor of 3,390.","quote":"NVIDIA Parakeet TDT 0.6B v2 Yes 6.05 3390 FastConformer","where_in_source":"Table 3","fact_url":"https://yapapp.cc/facts/open-asr-parakeet-tdt-v2-6-05-rtfx-3390","source_title":"Open ASR Leaderboard: Towards Reproducible and Transparent Multilingual and Long-Form Speech Recognition Evaluation","source_publisher":"arXiv","source_year":"2025","source_url":"https://arxiv.org/html/2510.06961v4","source_doi":null,"checked":"2026-10-16"},{"row":"HojoAI/Hojo-ASR-V1","value":6.065714285714286,"low":null,"high":null,"as_written":"6.07","unit":"percent","system":"HojoAI/Hojo-ASR-V1","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: apache-2.0): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, HojoAI/Hojo-ASR-V1 had an average word error rate of 6.07 percent.","quote":"HojoAI/Hojo-ASR-V1,6.065714285714286,73.39","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-hojoai-hojo-asr-v1-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini, CV German","value":6.07,"low":null,"high":null,"as_written":"6.07","unit":"percent","system":"Voxtral Mini","kind":"Open model","benchmark":"Common Voice","language":"German","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, Voxtral Mini had a word error rate of 6.07 percent on Common Voice German.","quote":"Model ar nl en fr de hi it pt es ... Voxtral Mini 63.98 6.03 10.22 8.92 6.07 12.74 5.91 7.76 4.98","where_in_source":"Table 5 (Common Voice), column de","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-common-voice-de-6-07","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"TheStageAI/thewhisper-large-v3-turbo","value":6.074285714285714,"low":null,"high":null,"as_written":"6.07","unit":"percent","system":"TheStageAI/thewhisper-large-v3-turbo","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: cc-by-4.0): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, TheStageAI/thewhisper-large-v3-turbo had an average word error rate of 6.07 percent.","quote":"TheStageAI/thewhisper-large-v3-turbo,6.074285714285714,1183.92","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-thestageai-thewhisper-large-v3-turbo-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, VoxPopuli English","value":6.14,"low":null,"high":null,"as_written":"6.14","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"VoxPopuli","language":"English","condition":"Mixed benchmark","year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 6.14 percent on the VoxPopuli test set in the Open ASR Leaderboard table.","quote":"| **Model** | **Avg WER** | **AMI** | **Earnings-22** | **GigaSpeech** | **LS test-clean** | **LS test-other** | **SPGI Speech** | **TEDLIUM-v3** | **VoxPopuli** | ... | `parakeet-tdt-0.6b-v3` | 6.34% | 11.31% | 11.42% | 9.59% | 1.93% | 3.59% | 3.97% | 2.75% | 6.14% |","where_in_source":"Open-ASR-Leaderboard table, VoxPopuli column","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-voxpopuli-open-asr-6-14","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Canary 1B Flash, CV French","value":6.15,"low":null,"high":null,"as_written":"6.15","unit":"percent","system":"NVIDIA Canary 1B Flash","kind":"Open model","benchmark":"Common Voice","language":"French","condition":null,"year":null,"conversion":null,"claim":"NVIDIA Canary 1B Flash had a word error rate of 6.15 percent on Common Voice French.","quote":"WER on [MCV-16.1](https://commonvoice.mozilla.org/en/datasets) test set: | **Version** | **Model**  |  **En**   | **De**   | **Es**   | **Fr**   | ... | nemo-main   | canary-1b-flash | 6.99 | 4.09 | 3.62 | 6.15 |","where_in_source":"WER table, fr column","fact_url":"https://yapapp.cc/facts/nvidia-canary-1b-flash-mcv-fr-6-15","source_title":"canary-1b-flash model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"n.d.","source_url":"https://huggingface.co/nvidia/canary-1b-flash/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Small, CV French","value":6.18,"low":null,"high":null,"as_written":"6.18","unit":"percent","system":"Voxtral Small","kind":"Open model","benchmark":"Common Voice","language":"French","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, Voxtral Small had a word error rate of 6.18 percent on Common Voice French.","quote":"Model ar nl en fr de hi it pt es ... Voxtral Small 61.97 3.98 8.58 6.18 3.74 9.01 3.96 6.43 3.31","where_in_source":"Table 5 (Common Voice), column fr","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-small-common-voice-fr-6-18","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"bosonai/higgs-audio-v3-stt","value":6.1899999999999995,"low":null,"high":null,"as_written":"6.19","unit":"percent","system":"bosonai/higgs-audio-v3-stt","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: apache-2.0): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, bosonai/higgs-audio-v3-stt had an average word error rate of 6.19 percent.","quote":"bosonai/higgs-audio-v3-stt,6.1899999999999995,109.65","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-bosonai-higgs-audio-v3-stt-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"Canary 1B v2, VoxPopuli English","value":6.25,"low":null,"high":null,"as_written":"6.25","unit":"percent","system":"NVIDIA Canary 1B v2","kind":"Open model","benchmark":"VoxPopuli","language":"English","condition":"Mixed benchmark","year":2025,"conversion":null,"claim":"NVIDIA Canary 1B v2 had a word error rate of 6.25 percent on the VoxPopuli test set in the Open ASR Leaderboard table.","quote":"|  **WER ↓**   | **RTFx**   | **Mean**   | **AMI**   | **GigaSpeech**   | **LS Clean**   | **LS Other**   | **Earnings22**   | **SPGISpech**   | **Tedlium**   | **Voxpopuli**   | ... | `Canary-1b-v2` | 749 | 7.15 | 16.01 | 10.82 | 2.18 | 3.56 | 11.79 | 2.28 | 4.29 | 6.25 |","where_in_source":"Open ASR Leaderboard table, VoxPopuli column","fact_url":"https://yapapp.cc/facts/nvidia-canary-1b-v2-voxpopuli-open-asr-6-25","source_title":"canary-1b-v2 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/canary-1b-v2/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v3, CV German","value":6.25,"low":null,"high":null,"as_written":"6.25","unit":"percent","system":"Whisper large-v3","kind":"Open model","benchmark":"Common Voice","language":"German","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, Whisper large-v3 had a word error rate of 6.25 percent on Common Voice German.","quote":"Model ar nl en fr de hi it pt es ... Whisper large-v3 50.58 5.83 22.91 11.33 6.25 46.75 6.81 7.17 5.66","where_in_source":"Table 5 (Common Voice), column de","fact_url":"https://yapapp.cc/facts/voxtral-2025-whisper-large-v3-common-voice-de-6-25","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"CohereLabs/cohere-transcribe-03-2026","value":6.2628571428571425,"low":null,"high":null,"as_written":"6.26","unit":"percent","system":"CohereLabs/cohere-transcribe-03-2026","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: apache-2.0): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, CohereLabs/cohere-transcribe-03-2026 had an average word error rate of 6.26 percent.","quote":"CohereLabs/cohere-transcribe-03-2026,6.2628571428571425,923.71","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-coherelabs-cohere-transcribe-03-2026-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini, FLEURS Dutch","value":6.27,"low":null,"high":null,"as_written":"6.27","unit":"percent","system":"Voxtral Mini","kind":"Open model","benchmark":"FLEURS","language":"Dutch","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, Voxtral Mini had a word error rate of 6.27 percent on FLEURS Dutch.","quote":"Model ar nl en fr de hi it pt es ... Voxtral Mini 25.40 6.27 3.77 4.87 4.40 9.26 2.51 3.76 3.52","where_in_source":"Table 4 (FLEURS), column nl","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-fleurs-nl-6-27","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"OpenMOSS-Team/MOSS-Transcribe-Diarize","value":6.2871428571428565,"low":null,"high":null,"as_written":"6.29","unit":"percent","system":"OpenMOSS-Team/MOSS-Transcribe-Diarize","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: apache-2.0): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, OpenMOSS-Team/MOSS-Transcribe-Diarize had an average word error rate of 6.29 percent.","quote":"OpenMOSS-Team/MOSS-Transcribe-Diarize,6.2871428571428565,387.46","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-openmoss-team-moss-transcribe-diarize-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"Edge0/ARK-ASR-0.6B","value":6.291428571428571,"low":null,"high":null,"as_written":"6.29","unit":"percent","system":"Edge0/ARK-ASR-0.6B","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: apache-2.0): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, Edge0/ARK-ASR-0.6B had an average word error rate of 6.29 percent.","quote":"Edge0/ARK-ASR-0.6B,6.291428571428571,673.02","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-edge0-ark-asr-0-6b-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Small SFT, EN short-form avg","value":6.31,"low":null,"high":null,"as_written":"6.31","unit":"percent","system":"Voxtral Small SFT","kind":"Open model","benchmark":"English short-form average (Voxtral paper)","language":"English","condition":"Mixed benchmark","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 2, Voxtral Small SFT had a macro-average word error rate of 6.31 percent across the English short-form test sets.","quote":"Voxtral Small SFT 86.61 ± 0.96 4.16 ± 0.03 6.31","where_in_source":"Table 2, En Short WER column","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-small-sft-en-short-form-average-6-31","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"NVIDIA Parakeet TDT 0.6B v3","value":6.32,"low":null,"high":null,"as_written":"6.32","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2025,"conversion":null,"claim":"On the Open ASR Leaderboard's English short-form tests, NVIDIA Parakeet TDT 0.6B v3 had an average word error rate of 6.32 percent, with an RTFx of 3,330.","quote":"NVIDIA Parakeet TDT 0.6B v3 Yes 6.32 3330 FastConformer","where_in_source":"Table 3, short-form English","fact_url":"https://yapapp.cc/facts/open-asr-parakeet-tdt-v3-6-32-rtfx-3330","source_title":"Open ASR Leaderboard: Towards Reproducible and Transparent Multilingual and Long-Form Speech Recognition Evaluation","source_publisher":"arXiv","source_year":"2025","source_url":"https://arxiv.org/html/2510.06961v4","source_doi":null,"checked":"2026-10-17"},{"row":"Voxtral Small Offline DPO, English","value":6.32,"low":null,"high":null,"as_written":"6.32","unit":"percent","system":"Voxtral Small Offline DPO","kind":"Open model","benchmark":"English short-form average (Voxtral paper)","language":"English","condition":"Mixed benchmark","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 2, Voxtral Small Offline DPO had a macro-average word error rate of 6.32 percent across the English short-form test sets.","quote":"Voxtral Small Offline DPO 87.29 ± 1.65 4.19 ± 0.04 6.32","where_in_source":"Table 2, En Short WER column","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-small-offline-dpo-en-short-form-average-6-32","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, Open ASR avg English","value":6.34,"low":null,"high":null,"as_written":"6.34","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had an average word error rate of 6.34 percent across the Open ASR Leaderboard's English datasets.","quote":"| **Model** | **Avg WER** | **AMI** | **Earnings-22** | **GigaSpeech** | **LS test-clean** | **LS test-other** | **SPGI Speech** | **TEDLIUM-v3** | **VoxPopuli** | ... | `parakeet-tdt-0.6b-v3` | 6.34% | 11.31% | 11.42% | 9.59% | 1.93% | 3.59% | 3.97% | 2.75% | 6.14% |","where_in_source":"Open-ASR-Leaderboard table, Avg WER column","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-open-asr-average-6-34","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"nvidia/canary-qwen-2.5b","value":6.368571428571429,"low":null,"high":null,"as_written":"6.37","unit":"percent","system":"nvidia/canary-qwen-2.5b","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: cc-by-4.0): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, nvidia/canary-qwen-2.5b had an average word error rate of 6.37 percent.","quote":"nvidia/canary-qwen-2.5b,6.368571428571429,881.67","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-nvidia-canary-qwen-2-5b-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"nvidia/parakeet-tdt-0.6b-v3","value":6.377142857142857,"low":null,"high":null,"as_written":"6.38","unit":"percent","system":"nvidia/parakeet-tdt-0.6b-v3 (fast-gpu-asr)","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: cc-by-4.0): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, nvidia/parakeet-tdt-0.6b-v3 (fast-gpu-asr) had an average word error rate of 6.38 percent.","quote":"nvidia/parakeet-tdt-0.6b-v3 (fast-gpu-asr),6.377142857142857,13304.58","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-nvidia-parakeet-tdt-0-6b-v3-fast-gpu-asr-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"ibm-granite/granite-speech-5.0-470m-turb","value":6.407142857142857,"low":null,"high":null,"as_written":"6.41","unit":"percent","system":"ibm-granite/granite-speech-5.0-470m-turboctc-nc","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: cc-by-nc-sa-4.0): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, ibm-granite/granite-speech-5.0-470m-turboctc-nc had an average word error rate of 6.41 percent.","quote":"ibm-granite/granite-speech-5.0-470m-turboctc-nc,6.407142857142857,20801.33","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-ibm-granite-granite-speech-5-0-470m-turboctc-nc-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"ibm-granite/granite-speech-4.1-2b","value":6.421428571428571,"low":null,"high":null,"as_written":"6.42","unit":"percent","system":"ibm-granite/granite-speech-4.1-2b","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: apache-2.0): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, ibm-granite/granite-speech-4.1-2b had an average word error rate of 6.42 percent.","quote":"ibm-granite/granite-speech-4.1-2b,6.421428571428571,546.84","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-ibm-granite-granite-speech-4-1-2b-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Small, CV Portuguese","value":6.43,"low":null,"high":null,"as_written":"6.43","unit":"percent","system":"Voxtral Small","kind":"Open model","benchmark":"Common Voice","language":"Portuguese","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, Voxtral Small had a word error rate of 6.43 percent on Common Voice Portuguese.","quote":"Model ar nl en fr de hi it pt es ... Voxtral Small 61.97 3.98 8.58 6.18 3.74 9.01 3.96 6.43 3.31","where_in_source":"Table 5 (Common Voice), column pt","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-small-common-voice-pt-6-43","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"nvidia/parakeet-tdt-0.6b-v2","value":6.432857142857143,"low":null,"high":null,"as_written":"6.43","unit":"percent","system":"nvidia/parakeet-tdt-0.6b-v2","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: cc-by-4.0): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, nvidia/parakeet-tdt-0.6b-v2 had an average word error rate of 6.43 percent.","quote":"nvidia/parakeet-tdt-0.6b-v2,6.432857142857143,6148.32","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-nvidia-parakeet-tdt-0-6b-v2-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"nvidia/parakeet-tdt-0.6b-v2","value":6.437142857142858,"low":null,"high":null,"as_written":"6.44","unit":"percent","system":"nvidia/parakeet-tdt-0.6b-v2 (fast-gpu-asr)","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: cc-by-4.0): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, nvidia/parakeet-tdt-0.6b-v2 (fast-gpu-asr) had an average word error rate of 6.44 percent.","quote":"nvidia/parakeet-tdt-0.6b-v2 (fast-gpu-asr),6.437142857142858,14383.48","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-nvidia-parakeet-tdt-0-6b-v2-fast-gpu-asr-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"ibm-granite/granite-speech-4.1-2b-nar","value":6.45,"low":null,"high":null,"as_written":"6.45","unit":"percent","system":"ibm-granite/granite-speech-4.1-2b-nar","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: apache-2.0): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, ibm-granite/granite-speech-4.1-2b-nar had an average word error rate of 6.45 percent.","quote":"ibm-granite/granite-speech-4.1-2b-nar,6.45,2112.58","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-ibm-granite-granite-speech-4-1-2b-nar-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, CoVoST 2 Dutch","value":6.5,"low":null,"high":null,"as_written":"6.5","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"CoVoST 2","language":"Dutch","condition":null,"year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 6.50 percent on the CoVoST 2 Dutch test.","quote":"| Language | Fleurs | MLS | CoVoST | ... | **nl** | 7.48% | 12.78% | 6.50% |","where_in_source":"Multilingual ASR table, CoVoST column, nl row","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-covost-2-nl-6-50","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Small Online DPO, English","value":6.5,"low":null,"high":null,"as_written":"6.5","unit":"percent","system":"Voxtral Small Online DPO","kind":"Open model","benchmark":"English short-form average (Voxtral paper)","language":"English","condition":"Mixed benchmark","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 2, Voxtral Small Online DPO had a macro-average word error rate of 6.50 percent across the English short-form test sets.","quote":"Voxtral Small Online DPO 88.31 ± 2.03 4.38 ± 0.06 6.50","where_in_source":"Table 2, En Short WER column","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-small-online-dpo-en-short-form-average-6-50","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"nvidia/parakeet-tdt-0.6b-v3","value":6.514285714285713,"low":null,"high":null,"as_written":"6.51","unit":"percent","system":"nvidia/parakeet-tdt-0.6b-v3","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: cc-by-4.0): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, nvidia/parakeet-tdt-0.6b-v3 had an average word error rate of 6.51 percent.","quote":"nvidia/parakeet-tdt-0.6b-v3,6.514285714285713,6227.73","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-nvidia-parakeet-tdt-0-6b-v3-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"Edge0/ARK-ASR-3B","value":6.525714285714286,"low":null,"high":null,"as_written":"6.53","unit":"percent","system":"Edge0/ARK-ASR-3B","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: apache-2.0): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, Edge0/ARK-ASR-3B had an average word error rate of 6.53 percent.","quote":"Edge0/ARK-ASR-3B,6.525714285714286,480.07","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-edge0-ark-asr-3b-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"Canary 1B, CV French","value":6.53,"low":null,"high":null,"as_written":"6.53","unit":"percent","system":"NVIDIA Canary 1B","kind":"Open model","benchmark":"Common Voice","language":"French","condition":null,"year":null,"conversion":null,"claim":"NVIDIA Canary 1B had a word error rate of 6.53 percent on Common Voice French.","quote":"WER on [MCV-16.1](https://commonvoice.mozilla.org/en/datasets) test set: | **Version** | **Model**     | **En**   | **De**   | **Es**   | **Fr**   | ... | 1.23.0  | canary-1b | 7.97 | 4.61 | 3.99 | 6.53 |","where_in_source":"WER table, fr column","fact_url":"https://yapapp.cc/facts/nvidia-canary-1b-mcv-fr-6-53","source_title":"canary-1b model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"n.d.","source_url":"https://huggingface.co/nvidia/canary-1b/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Qwen/Qwen3-ASR-0.6B-hf","value":6.602857142857144,"low":null,"high":null,"as_written":"6.6","unit":"percent","system":"Qwen/Qwen3-ASR-0.6B-hf","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: apache-2.0): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, Qwen/Qwen3-ASR-0.6B-hf had an average word error rate of 6.60 percent.","quote":"Qwen/Qwen3-ASR-0.6B-hf,6.602857142857144,749.52","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-qwen-qwen3-asr-0-6b-hf-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v3-turbo, French (our test)","value":6.62,"low":null,"high":null,"as_written":"6.62","unit":"percent","system":"Whisper large-v3-turbo","kind":"Open model","benchmark":"FLEURS","language":"French","condition":"Clean read speech","year":2026,"conversion":null,"claim":"In our October 2026 benchmark on a Mac with an M2 Pro, Whisper large-v3-turbo had a word error rate of 6.62 percent on 100 French recordings from the FLEURS test set.","quote":null,"where_in_source":null,"fact_url":"https://yapapp.cc/facts/m-fleurs-2026-10-whisper-fr-wer","source_title":null,"source_publisher":null,"source_year":null,"source_url":null,"source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Small, VoxPopuli English","value":6.62,"low":null,"high":null,"as_written":"6.62","unit":"percent","system":"Voxtral Small","kind":"Open model","benchmark":"VoxPopuli","language":"English","condition":"Mixed benchmark","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, Voxtral Small had a word error rate of 6.62 percent on VoxPopuli.","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... Voxtral Small 1.53 3.14 10.27 6.62 11.09 9.64 1.89 9.55 12.48","where_in_source":"Table 3 (short-form English), VoxPopuli column","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-small-voxpopuli-6-62","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini Transcribe, CV Portuguese","value":6.7,"low":null,"high":null,"as_written":"6.7","unit":"percent","system":"Voxtral Mini Transcribe","kind":"Open model","benchmark":"Common Voice","language":"Portuguese","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, Voxtral Mini Transcribe had a word error rate of 6.70 percent on Common Voice Portuguese.","quote":"Model ar nl en fr de hi it pt es ... Voxtral Mini Transcribe 62.01 4.71 8.25 7.29 4.85 10.42 4.37 6.70 3.96","where_in_source":"Table 5 (Common Voice), column pt","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-transcribe-common-voice-pt-6-70","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v2, Dutch","value":6.7,"low":null,"high":null,"as_written":"6.7","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Dutch","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS read-speech benchmark, Whisper large-v2 had a word error rate of 6.7 percent in Dutch.","quote":"Whisper large-v2 6.7 75.3 102.4 5.4 93.7 4.3 14.4 5.6","where_in_source":"Appendix D.2.4, FLEURS transcription table, Dutch column","fact_url":"https://yapapp.cc/facts/whisper-fleurs-dutch-6-7","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-17"},{"row":"Voxtral Mini, MLS Portuguese","value":6.72,"low":null,"high":null,"as_written":"6.72","unit":"percent","system":"Voxtral Mini","kind":"Open model","benchmark":"Multilingual LibriSpeech","language":"Portuguese","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 6, Voxtral Mini had a word error rate of 6.72 percent on Multilingual LibriSpeech Portuguese.","quote":"Model nl fr de it pt es ... Voxtral Mini 10.09 5.28 7.09 11.30 6.72 5.12","where_in_source":"Table 6 (Multilingual LibriSpeech), column pt","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-multilingual-librispeech-pt-6-72","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini SFT, EN short-form avg","value":6.77,"low":null,"high":null,"as_written":"6.77","unit":"percent","system":"Voxtral Mini SFT","kind":"Open model","benchmark":"English short-form average (Voxtral paper)","language":"English","condition":"Mixed benchmark","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 2, Voxtral Mini SFT had a macro-average word error rate of 6.77 percent across the English short-form test sets.","quote":"Voxtral Mini SFT 83.47 ± 2.17 3.92 ± 0.04 6.77","where_in_source":"Table 2, En Short WER column","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-sft-en-short-form-average-6-77","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini Offline DPO, English","value":6.78,"low":null,"high":null,"as_written":"6.78","unit":"percent","system":"Voxtral Mini Offline DPO","kind":"Open model","benchmark":"English short-form average (Voxtral paper)","language":"English","condition":"Mixed benchmark","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 2, Voxtral Mini Offline DPO had a macro-average word error rate of 6.78 percent across the English short-form test sets.","quote":"Voxtral Mini Offline DPO 84.91 ± 3.21 3.92 ± 0.08 6.78","where_in_source":"Table 2, En Short WER column","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-offline-dpo-en-short-form-average-6-78","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, FLEURS Ukrainian","value":6.79,"low":null,"high":null,"as_written":"6.79","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"FLEURS","language":"Ukrainian","condition":"Clean read speech","year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 6.79 percent on the FLEURS Ukrainian test.","quote":"| Language | Fleurs | MLS | CoVoST | ... | **uk** | 6.79% | - | 5.10% |","where_in_source":"Multilingual ASR table, Fleurs column, uk row","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-fleurs-uk-6-79","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini Online DPO, English","value":6.79,"low":null,"high":null,"as_written":"6.79","unit":"percent","system":"Voxtral Mini Online DPO","kind":"Open model","benchmark":"English short-form average (Voxtral paper)","language":"English","condition":"Mixed benchmark","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 2, Voxtral Mini Online DPO had a macro-average word error rate of 6.79 percent across the English short-form test sets.","quote":"Voxtral Mini Online DPO 85.59 ± 3.77 4.08 ± 0.07 6.79","where_in_source":"Table 2, En Short WER column","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-online-dpo-en-short-form-average-6-79","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, CoVoST 2 English","value":6.8,"low":null,"high":null,"as_written":"6.8","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"CoVoST 2","language":"English","condition":null,"year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 6.80 percent on the CoVoST 2 English test.","quote":"| Language | Fleurs | MLS | CoVoST | ... | **en** | 4.85% | - | 6.80% |","where_in_source":"Multilingual ASR table, CoVoST column, en row","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-covost-2-en-6-80","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v3, CV Italian","value":6.81,"low":null,"high":null,"as_written":"6.81","unit":"percent","system":"Whisper large-v3","kind":"Open model","benchmark":"Common Voice","language":"Italian","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, Whisper large-v3 had a word error rate of 6.81 percent on Common Voice Italian.","quote":"Model ar nl en fr de hi it pt es ... Whisper large-v3 50.58 5.83 22.91 11.33 6.25 46.75 6.81 7.17 5.66","where_in_source":"Table 5 (Common Voice), column it","fact_url":"https://yapapp.cc/facts/voxtral-2025-whisper-large-v3-common-voice-it-6-81","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v3-turbo, Ukrainian (our test)","value":6.83,"low":null,"high":null,"as_written":"6.83","unit":"percent","system":"Whisper large-v3-turbo","kind":"Open model","benchmark":"FLEURS","language":"Ukrainian","condition":"Clean read speech","year":2026,"conversion":null,"claim":"In our October 2026 benchmark on a Mac with an M2 Pro, Whisper large-v3-turbo had a word error rate of 6.83 percent on 100 Ukrainian recordings from the FLEURS test set.","quote":null,"where_in_source":null,"fact_url":"https://yapapp.cc/facts/m-fleurs-2026-10-whisper-uk-wer","source_title":null,"source_publisher":null,"source_year":null,"source_url":null,"source_doi":null,"checked":"2026-10-19"},{"row":"ibm-granite/granite-speech-5.0-470m-turb","value":6.837142857142857,"low":null,"high":null,"as_written":"6.84","unit":"percent","system":"ibm-granite/granite-speech-5.0-470m-turboctc","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: apache-2.0): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, ibm-granite/granite-speech-5.0-470m-turboctc had an average word error rate of 6.84 percent.","quote":"ibm-granite/granite-speech-5.0-470m-turboctc,6.837142857142857,21134.31","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-ibm-granite-granite-speech-5-0-470m-turboctc-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini, VoxPopuli English","value":6.85,"low":null,"high":null,"as_written":"6.85","unit":"percent","system":"Voxtral Mini","kind":"Open model","benchmark":"VoxPopuli","language":"English","condition":"Mixed benchmark","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, Voxtral Mini had a word error rate of 6.85 percent on VoxPopuli.","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... Voxtral Mini 1.86 4.04 10.68 6.85 11.32 10.59 2.19 9.62 11.84","where_in_source":"Table 3 (short-form English), VoxPopuli column","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-voxpopuli-6-85","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"nvidia/nemotron-speech-streaming-en-0.6b","value":6.894285714285714,"low":null,"high":null,"as_written":"6.89","unit":"percent","system":"nvidia/nemotron-speech-streaming-en-0.6b","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: nvidia-open-model-license): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, nvidia/nemotron-speech-streaming-en-0.6b had an average word error rate of 6.89 percent.","quote":"nvidia/nemotron-speech-streaming-en-0.6b,6.894285714285714,1114.14","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-nvidia-nemotron-speech-streaming-en-0-6b-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"mistralai/Voxtral-Small-24B-2507","value":6.895714285714285,"low":null,"high":null,"as_written":"6.9","unit":"percent","system":"mistralai/Voxtral-Small-24B-2507","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: apache-2.0): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, mistralai/Voxtral-Small-24B-2507 had an average word error rate of 6.90 percent.","quote":"mistralai/Voxtral-Small-24B-2507,6.895714285714285,101.45","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-mistralai-voxtral-small-24b-2507-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"Canary 1B Flash, CV English","value":6.99,"low":null,"high":null,"as_written":"6.99","unit":"percent","system":"NVIDIA Canary 1B Flash","kind":"Open model","benchmark":"Common Voice","language":"English","condition":null,"year":null,"conversion":null,"claim":"NVIDIA Canary 1B Flash had a word error rate of 6.99 percent on Common Voice English.","quote":"WER on [MCV-16.1](https://commonvoice.mozilla.org/en/datasets) test set: | **Version** | **Model**  |  **En**   | **De**   | **Es**   | **Fr**   | ... | nemo-main   | canary-1b-flash | 6.99 | 4.09 | 3.62 | 6.15 |","where_in_source":"WER table, en column","fact_url":"https://yapapp.cc/facts/nvidia-canary-1b-flash-mcv-en-6-99","source_title":"canary-1b-flash model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"n.d.","source_url":"https://huggingface.co/nvidia/canary-1b-flash/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"ibm-granite/granite-4.0-1b-speech","value":7.0200000000000005,"low":null,"high":null,"as_written":"7.02","unit":"percent","system":"ibm-granite/granite-4.0-1b-speech","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: apache-2.0): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, ibm-granite/granite-4.0-1b-speech had an average word error rate of 7.02 percent.","quote":"ibm-granite/granite-4.0-1b-speech,7.0200000000000005,666.99","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-ibm-granite-granite-4-0-1b-speech-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"distil-whisper/distil-large-v3.5","value":7.022857142857143,"low":null,"high":null,"as_written":"7.02","unit":"percent","system":"distil-whisper/distil-large-v3.5","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: mit): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, distil-whisper/distil-large-v3.5 had an average word error rate of 7.02 percent.","quote":"distil-whisper/distil-large-v3.5,7.022857142857143,898.21","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-distil-whisper-distil-large-v3-5-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v3, MLS Portuguese","value":7.03,"low":null,"high":null,"as_written":"7.03","unit":"percent","system":"Whisper large-v3","kind":"Open model","benchmark":"Multilingual LibriSpeech","language":"Portuguese","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 6, Whisper large-v3 had a word error rate of 7.03 percent on Multilingual LibriSpeech Portuguese.","quote":"Model nl fr de it pt es ... Whisper large-v3 9.19 5.09 5.72 9.78 7.03 3.89","where_in_source":"Table 6 (Multilingual LibriSpeech), column pt","fact_url":"https://yapapp.cc/facts/voxtral-2025-whisper-large-v3-multilingual-librispeech-pt-7-03","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini, MLS German","value":7.09,"low":null,"high":null,"as_written":"7.09","unit":"percent","system":"Voxtral Mini","kind":"Open model","benchmark":"Multilingual LibriSpeech","language":"German","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 6, Voxtral Mini had a word error rate of 7.09 percent on Multilingual LibriSpeech German.","quote":"Model nl fr de it pt es ... Voxtral Mini 10.09 5.28 7.09 11.30 6.72 5.12","where_in_source":"Table 6 (Multilingual LibriSpeech), column de","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-multilingual-librispeech-de-7-09","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v2, Indonesian","value":7.1,"low":null,"high":null,"as_written":"7.1","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Indonesian","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 7.1 percent for Indonesian.","quote":"Whisper large-v2 13.4 17.0 44.6 7.1 38.2 4.0 5.3 nan 105.0 37.7 99.7 37.0 14.3 88.0","where_in_source":"Table 13 (FLEURS), Whisper large-v2 row, column Indonesian","fact_url":"https://yapapp.cc/facts/whisper-fleurs-indonesian-7-1","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-19"},{"row":"Parakeet TDT v3, noise 10 dB","value":7.12,"low":null,"high":null,"as_written":"7.12","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"Open ASR sets with MUSAN noise (average)","language":"English","condition":"Noisy or distant","year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had an average word error rate of 7.12 percent on the Open ASR sets with MUSAN noise at 10 dB SNR.","quote":"| **SNR Level** | **Avg WER** | **AMI** | **Earnings** | **GigaSpeech** | **LS test-clean** | **LS test-other** | **SPGI** | **Tedlium** | **VoxPopuli** | **Relative Change** | ... | SNR 10 | 7.12% | 13.99% | 11.79% | 9.96% | 2.15% | 4.55% | 4.45% | 3.05% | 6.99% | -12.28% |","where_in_source":"Noise robustness table, SNR 10 row, Avg WER column","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-musan-snr-10-db-7-12","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Canary 1B v2, Open ASR avg","value":7.15,"low":null,"high":null,"as_written":"7.15","unit":"percent","system":"NVIDIA Canary 1B v2","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2025,"conversion":null,"claim":"On the Open ASR Leaderboard's English short-form tests, NVIDIA Canary 1B v2 had an average word error rate of 7.15 percent, with an RTFx of 749.","quote":"NVIDIA Canary 1B v2 Yes 7.15 749 FastConformer","where_in_source":"Table 3, short-form English","fact_url":"https://yapapp.cc/facts/open-asr-canary-1b-v2-7-15-rtfx-749","source_title":"Open ASR Leaderboard: Towards Reproducible and Transparent Multilingual and Long-Form Speech Recognition Evaluation","source_publisher":"arXiv","source_year":"2025","source_url":"https://arxiv.org/html/2510.06961v4","source_doi":null,"checked":"2026-10-17"},{"row":"Whisper large-v3, CV Portuguese","value":7.17,"low":null,"high":null,"as_written":"7.17","unit":"percent","system":"Whisper large-v3","kind":"Open model","benchmark":"Common Voice","language":"Portuguese","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, Whisper large-v3 had a word error rate of 7.17 percent on Common Voice Portuguese.","quote":"Model ar nl en fr de hi it pt es ... Whisper large-v3 50.58 5.83 22.91 11.33 6.25 46.75 6.81 7.17 5.66","where_in_source":"Table 5 (Common Voice), column pt","fact_url":"https://yapapp.cc/facts/voxtral-2025-whisper-large-v3-common-voice-pt-7-17","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large, Svarah","value":7.2,"low":null,"high":null,"as_written":"7.2","unit":"percent","system":"Whisper large","kind":"Open model","benchmark":"Svarah","language":"English","condition":"Accented or dialect","year":2023,"conversion":null,"claim":"On Svarah, a 2023 benchmark of English from Indian speakers, the large version of OpenAI's Whisper had a word error rate of 7.2 percent in the Svarah column.","quote":"Whisper large 1550M 7.2 1.6 2.7","where_in_source":"Table 1","fact_url":"https://yapapp.cc/facts/svarah-2023-whisper-large-7-2-wer","source_title":"Svarah: Evaluating English ASR Systems on Indian Accents","source_publisher":"arXiv","source_year":"2023","source_url":"https://arxiv.org/html/2305.15760","source_doi":null,"checked":"2026-10-17"},{"row":"Distil-Whisper Large v3.5","value":7.21,"low":null,"high":null,"as_written":"7.21","unit":"percent","system":"Distil-Whisper Large v3.5","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2025,"conversion":null,"claim":"On the Open ASR Leaderboard's English short-form tests, Distil-Whisper Large v3.5 had an average word error rate of 7.21 percent, with an RTFx of 202.","quote":"Distil-Whisper Large v3.5 Yes 7.21 202 Whisper","where_in_source":"Table 3, short-form English","fact_url":"https://yapapp.cc/facts/open-asr-distil-whisper-large-v3-5-7-21-rtfx-202","source_title":"Open ASR Leaderboard: Towards Reproducible and Transparent Multilingual and Long-Form Speech Recognition Evaluation","source_publisher":"arXiv","source_year":"2025","source_url":"https://arxiv.org/html/2510.06961v4","source_doi":null,"checked":"2026-10-17"},{"row":"FermionResearch/Phonon-2","value":7.228571428571429,"low":null,"high":null,"as_written":"7.23","unit":"percent","system":"FermionResearch/Phonon-2","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: cc-by-4.0): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, FermionResearch/Phonon-2 had an average word error rate of 7.23 percent.","quote":"FermionResearch/Phonon-2,7.228571428571429,1976.52","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-fermionresearch-phonon-2-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"Canary 1B v2, MLS","value":7.27,"low":null,"high":null,"as_written":"7.27","unit":"percent","system":"NVIDIA Canary 1B v2","kind":"Open model","benchmark":"Multilingual LibriSpeech","language":"European languages (average)","condition":"Clean read speech","year":2025,"conversion":null,"claim":"NVIDIA Canary 1B v2 had an average word error rate of 7.27 percent on the Multilingual LibriSpeech test across its MLS-6 languages.","quote":"| **WER ↓**          | Fleurs-25 Langs       | CoVoST-13 Langs      | MLS - 6 Langs      | ... | **`Canary-1b-v2`** | 8.40%                 | 8.85%                | 7.27%              |","where_in_source":"ASR table, MLS-6 column","fact_url":"https://yapapp.cc/facts/nvidia-canary-1b-v2-multilingual-librispeech-average-7-27","source_title":"canary-1b-v2 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/canary-1b-v2/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, MLS Polish","value":7.28,"low":null,"high":null,"as_written":"7.28","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"Multilingual LibriSpeech","language":"Polish","condition":"Clean read speech","year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 7.28 percent on the Multilingual LibriSpeech Polish test.","quote":"| Language | Fleurs | MLS | CoVoST | ... | **pl** | 7.31% | 7.28% | - |","where_in_source":"Multilingual ASR table, MLS column, pl row","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-multilingual-librispeech-pl-7-28","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini Transcribe, CV French","value":7.29,"low":null,"high":null,"as_written":"7.29","unit":"percent","system":"Voxtral Mini Transcribe","kind":"Open model","benchmark":"Common Voice","language":"French","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, Voxtral Mini Transcribe had a word error rate of 7.29 percent on Common Voice French.","quote":"Model ar nl en fr de hi it pt es ... Voxtral Mini Transcribe 62.01 4.71 8.25 7.29 4.85 10.42 4.37 6.70 3.96","where_in_source":"Table 5 (Common Voice), column fr","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-transcribe-common-voice-fr-7-29","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v2, Catalan","value":7.3,"low":null,"high":null,"as_written":"7.3","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Catalan","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 7.3 percent for Catalan.","quote":"Whisper large-v2 36.7 140.3 16.0 106.2 23.4 45.4 14.6 104.1 15.7 7.3 14.7 13.3 33.0 13.8","where_in_source":"Table 13 (FLEURS), Whisper large-v2 row, column Catalan","fact_url":"https://yapapp.cc/facts/whisper-fleurs-catalan-7-3","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-19"},{"row":"Parakeet TDT v3, FLEURS Polish","value":7.31,"low":null,"high":null,"as_written":"7.31","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"FLEURS","language":"Polish","condition":"Clean read speech","year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 7.31 percent on the FLEURS Polish test.","quote":"| Language | Fleurs | MLS | CoVoST | ... | **pl** | 7.31% | 7.28% | - |","where_in_source":"Multilingual ASR table, Fleurs column, pl row","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-fleurs-pl-7-31","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Zipformer XL, transducer beam search","value":7.381428571428572,"low":null,"high":null,"as_written":"7.38","unit":"percent","system":"soundsgoodai/Zipformer-cr-ctc-transducer-XL-290M (fast-gpu-asr, transducer_modified_beam_search)","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: cc-by-nc-4.0): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, soundsgoodai/Zipformer-cr-ctc-transducer-XL-290M (fast-gpu-asr, transducer_modified_beam_search) had an average word error rate of 7.38 percent.","quote":"transducer_modified_beam_search)\",7.381428571428572,19289.67","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-soundsgoodai-zipformer-cr-ctc-transducer-xl-290m-fast-gpu-asr-transducer-modified-beam-search-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"OpenAI Whisper Large v3","value":7.44,"low":null,"high":null,"as_written":"7.44","unit":"percent","system":"OpenAI Whisper Large v3","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2025,"conversion":null,"claim":"On the same leaderboard track, OpenAI Whisper Large v3 had an average word error rate of 7.44 percent and an inverse real-time factor of 146.","quote":"OpenAI Whisper Large v3 Yes 7.44 146 Whisper","where_in_source":"Table 3","fact_url":"https://yapapp.cc/facts/open-asr-whisper-large-v3-7-44-rtfx-146","source_title":"Open ASR Leaderboard: Towards Reproducible and Transparent Multilingual and Long-Form Speech Recognition Evaluation","source_publisher":"arXiv","source_year":"2025","source_url":"https://arxiv.org/html/2510.06961v4","source_doi":null,"checked":"2026-10-16"},{"row":"Parakeet TDT v3, FLEURS Dutch","value":7.48,"low":null,"high":null,"as_written":"7.48","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"FLEURS","language":"Dutch","condition":"Clean read speech","year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 7.48 percent on the FLEURS Dutch test.","quote":"| Language | Fleurs | MLS | CoVoST | ... | **nl** | 7.48% | 12.78% | 6.50% |","where_in_source":"Multilingual ASR table, Fleurs column, nl row","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-fleurs-nl-7-48","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, MLS Portuguese","value":7.5,"low":null,"high":null,"as_written":"7.5","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"Multilingual LibriSpeech","language":"Portuguese","condition":"Clean read speech","year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 7.50 percent on the Multilingual LibriSpeech Portuguese test.","quote":"| Language | Fleurs | MLS | CoVoST | ... | **pt** | 4.76% | 7.50% | 3.96% |","where_in_source":"Multilingual ASR table, MLS column, pt row","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-multilingual-librispeech-pt-7-50","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Small, FLEURS Hindi","value":7.69,"low":null,"high":null,"as_written":"7.69","unit":"percent","system":"Voxtral Small","kind":"Open model","benchmark":"FLEURS","language":"Hindi","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, Voxtral Small had a word error rate of 7.69 percent on FLEURS Hindi.","quote":"Model ar nl en fr de hi it pt es ... Voxtral Small 13.44 4.94 3.35 4.03 3.38 7.69 2.62 3.79 2.72","where_in_source":"Table 4 (FLEURS), column hi","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-small-fleurs-hi-7-69","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini, CV Portuguese","value":7.76,"low":null,"high":null,"as_written":"7.76","unit":"percent","system":"Voxtral Mini","kind":"Open model","benchmark":"Common Voice","language":"Portuguese","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, Voxtral Mini had a word error rate of 7.76 percent on Common Voice Portuguese.","quote":"Model ar nl en fr de hi it pt es ... Voxtral Mini 63.98 6.03 10.22 8.92 6.07 12.74 5.91 7.76 4.98","where_in_source":"Table 5 (Common Voice), column pt","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-common-voice-pt-7-76","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"FunAudioLLM/Fun-ASR-Nano-2512-hf","value":7.761428571428572,"low":null,"high":null,"as_written":"7.76","unit":"percent","system":"FunAudioLLM/Fun-ASR-Nano-2512-hf","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: apache-2.0): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, FunAudioLLM/Fun-ASR-Nano-2512-hf had an average word error rate of 7.76 percent.","quote":"FunAudioLLM/Fun-ASR-Nano-2512-hf,7.761428571428572,506.37","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-funaudiollm-fun-asr-nano-2512-hf-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, MLS","value":7.83,"low":null,"high":null,"as_written":"7.83","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"Multilingual LibriSpeech","language":"European languages (average)","condition":"Clean read speech","year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had an average word error rate of 7.83 percent on the Multilingual LibriSpeech test across its European languages.","quote":"| Language | Fleurs | MLS | CoVoST | ... | **Average WER ↓** | *11.97%* | *7.83%* | *11.98%* |","where_in_source":"Multilingual ASR table, MLS column, average row","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-multilingual-librispeech-average-7-83","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"OpenAI Whisper Large v3 Turbo","value":7.83,"low":null,"high":null,"as_written":"7.83","unit":"percent","system":"OpenAI Whisper Large v3 Turbo","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2025,"conversion":null,"claim":"On the Open ASR Leaderboard's English short-form tests, OpenAI Whisper Large v3 Turbo had an average word error rate of 7.83 percent, with an RTFx of 200.","quote":"OpenAI Whisper Large v3 Turbo Yes 7.83 200","where_in_source":"Table 3, short-form English","fact_url":"https://yapapp.cc/facts/open-asr-whisper-large-v3-turbo-7-83-rtfx-200","source_title":"Open ASR Leaderboard: Towards Reproducible and Transparent Multilingual and Long-Form Speech Recognition Evaluation","source_publisher":"arXiv","source_year":"2025","source_url":"https://arxiv.org/html/2510.06961v4","source_doi":null,"checked":"2026-10-17"},{"row":"mistralai/Voxtral-Mini-3B-2507","value":7.837142857142856,"low":null,"high":null,"as_written":"7.84","unit":"percent","system":"mistralai/Voxtral-Mini-3B-2507","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: apache-2.0): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, mistralai/Voxtral-Mini-3B-2507 had an average word error rate of 7.84 percent.","quote":"mistralai/Voxtral-Mini-3B-2507,7.837142857142856,179.78","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-mistralai-voxtral-mini-3b-2507-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"openai/whisper-large-v3","value":7.841428571428572,"low":null,"high":null,"as_written":"7.84","unit":"percent","system":"openai/whisper-large-v3","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: apache-2.0): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, openai/whisper-large-v3 had an average word error rate of 7.84 percent.","quote":"openai/whisper-large-v3,7.841428571428572,470.73","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-openai-whisper-large-v3-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"nvidia/canary-1b-v2","value":7.911428571428572,"low":null,"high":null,"as_written":"7.91","unit":"percent","system":"nvidia/canary-1b-v2","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: cc-by-4.0): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, nvidia/canary-1b-v2 had an average word error rate of 7.91 percent.","quote":"nvidia/canary-1b-v2,7.911428571428572,1853.68","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-nvidia-canary-1b-v2-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"Canary 1B, CV English","value":7.97,"low":null,"high":null,"as_written":"7.97","unit":"percent","system":"NVIDIA Canary 1B","kind":"Open model","benchmark":"Common Voice","language":"English","condition":null,"year":null,"conversion":null,"claim":"NVIDIA Canary 1B had a word error rate of 7.97 percent on Common Voice English.","quote":"WER on [MCV-16.1](https://commonvoice.mozilla.org/en/datasets) test set: | **Version** | **Model**     | **En**   | **De**   | **Es**   | **Fr**   | ... | 1.23.0  | canary-1b | 7.97 | 4.61 | 3.99 | 6.53 |","where_in_source":"WER table, en column","fact_url":"https://yapapp.cc/facts/nvidia-canary-1b-mcv-en-7-97","source_title":"canary-1b model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"n.d.","source_url":"https://huggingface.co/nvidia/canary-1b/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"nvidia/parakeet-ctc-1.1b","value":8.215714285714288,"low":null,"high":null,"as_written":"8.22","unit":"percent","system":"nvidia/parakeet-ctc-1.1b","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: cc-by-4.0): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, nvidia/parakeet-ctc-1.1b had an average word error rate of 8.22 percent.","quote":"nvidia/parakeet-ctc-1.1b,8.215714285714288,5137.41","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-nvidia-parakeet-ctc-1-1b-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"microsoft/VibeVoice-ASR-HF","value":8.227142857142857,"low":null,"high":null,"as_written":"8.23","unit":"percent","system":"microsoft/VibeVoice-ASR-HF","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: mit): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, microsoft/VibeVoice-ASR-HF had an average word error rate of 8.23 percent.","quote":"microsoft/VibeVoice-ASR-HF,8.227142857142857,217.89","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-microsoft-vibevoice-asr-hf-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, noise 5 dB","value":8.23,"low":null,"high":null,"as_written":"8.23","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"Open ASR sets with MUSAN noise (average)","language":"English","condition":"Noisy or distant","year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had an average word error rate of 8.23 percent on the Open ASR sets with MUSAN noise at 5 dB SNR.","quote":"| **SNR Level** | **Avg WER** | **AMI** | **Earnings** | **GigaSpeech** | **LS test-clean** | **LS test-other** | **SPGI** | **Tedlium** | **VoxPopuli** | **Relative Change** | ... | SNR 5 | 8.23% | 17.59% | 13.01% | 10.69% | 2.62% | 6.05% | 5.23% | 3.33% | 7.31% | -29.81% |","where_in_source":"Noise robustness table, SNR 5 row, Avg WER column","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-musan-snr-5-db-8-23","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini Transcribe, CV English","value":8.25,"low":null,"high":null,"as_written":"8.25","unit":"percent","system":"Voxtral Mini Transcribe","kind":"Open model","benchmark":"Common Voice","language":"English","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, Voxtral Mini Transcribe had a word error rate of 8.25 percent on Common Voice English.","quote":"Model ar nl en fr de hi it pt es ... Voxtral Mini Transcribe 62.01 4.71 8.25 7.29 4.85 10.42 4.37 6.70 3.96","where_in_source":"Table 5 (Common Voice), column en","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-transcribe-common-voice-en-8-25","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v2, French","value":8.3,"low":null,"high":null,"as_written":"8.3","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"French","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the Fleurs benchmark, the Whisper large-v2 model had a word error rate of 8.3 percent for French.","quote":"Model German Greek English Spanish Estonian Persian Finnish Tagalog French Galician Gujarati Hausa Hebrew Hindi ... Whisper large-v2 4.5 12.5 4.2 3.0 21.9 32.9 9.7 13.8 8.3 15.4 102.7 88.9 27.1 21.5","where_in_source":"Table 13 (Fleurs)","fact_url":"https://yapapp.cc/facts/whisper-fleurs-french-8-3","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-16"},{"row":"openai/whisper-large-v3-turbo","value":8.387142857142857,"low":null,"high":null,"as_written":"8.39","unit":"percent","system":"openai/whisper-large-v3-turbo","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: mit): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, openai/whisper-large-v3-turbo had an average word error rate of 8.39 percent.","quote":"openai/whisper-large-v3-turbo,8.387142857142857,810.5","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-openai-whisper-large-v3-turbo-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"Canary 1B v2, FLEURS","value":8.4,"low":null,"high":null,"as_written":"8.4","unit":"percent","system":"NVIDIA Canary 1B v2","kind":"Open model","benchmark":"FLEURS","language":"European languages (average)","condition":"Clean read speech","year":2025,"conversion":null,"claim":"NVIDIA Canary 1B v2 had an average word error rate of 8.40 percent on the FLEURS test across its Fleurs-25 languages.","quote":"| **WER ↓**          | Fleurs-25 Langs       | CoVoST-13 Langs      | MLS - 6 Langs      | ... | **`Canary-1b-v2`** | 8.40%                 | 8.85%                | 7.27%              |","where_in_source":"ASR table, Fleurs-25 column","fact_url":"https://yapapp.cc/facts/nvidia-canary-1b-v2-fleurs-average-8-40","source_title":"canary-1b-v2 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/canary-1b-v2/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v2, Turkish","value":8.4,"low":null,"high":null,"as_written":"8.4","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Turkish","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 8.4 percent for Turkish.","quote":"Whisper large-v2 8.5 39.3 17.5 99.0 85.8 11.5 8.4 8.6 22.6 90.2 10.3 94.8","where_in_source":"Table 13 (FLEURS), Whisper large-v2 row, column Turkish","fact_url":"https://yapapp.cc/facts/whisper-fleurs-turkish-8-4","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-19"},{"row":"Voxtral Small, MLS Italian","value":8.44,"low":null,"high":null,"as_written":"8.44","unit":"percent","system":"Voxtral Small","kind":"Open model","benchmark":"Multilingual LibriSpeech","language":"Italian","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 6, Voxtral Small had a word error rate of 8.44 percent on Multilingual LibriSpeech Italian.","quote":"Model nl fr de it pt es ... Voxtral Small 9.43 3.73 5.57 8.44 5.85 3.62","where_in_source":"Table 6 (Multilingual LibriSpeech), column it","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-small-multilingual-librispeech-it-8-44","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v2, Swedish","value":8.5,"low":null,"high":null,"as_written":"8.5","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Swedish","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 8.5 percent for Swedish.","quote":"Whisper large-v2 8.5 39.3 17.5 99.0 85.8 11.5 8.4 8.6 22.6 90.2 10.3 94.8","where_in_source":"Table 13 (FLEURS), Whisper large-v2 row, column Swedish","fact_url":"https://yapapp.cc/facts/whisper-fleurs-swedish-8-5","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-19"},{"row":"espnet/owsm_ctc_v4_1B","value":8.554285714285713,"low":null,"high":null,"as_written":"8.55","unit":"percent","system":"espnet/owsm_ctc_v4_1B","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: cc-by-4.0): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, espnet/owsm_ctc_v4_1B had an average word error rate of 8.55 percent.","quote":"espnet/owsm_ctc_v4_1B,8.554285714285713,783.3","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-espnet-owsm-ctc-v4-1b-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Small, CV English","value":8.58,"low":null,"high":null,"as_written":"8.58","unit":"percent","system":"Voxtral Small","kind":"Open model","benchmark":"Common Voice","language":"English","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, Voxtral Small had a word error rate of 8.58 percent on Common Voice English.","quote":"Model ar nl en fr de hi it pt es ... Voxtral Small 61.97 3.98 8.58 6.18 3.74 9.01 3.96 6.43 3.31","where_in_source":"Table 5 (Common Voice), column en","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-small-common-voice-en-8-58","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v2, Ukrainian","value":8.6,"low":null,"high":null,"as_written":"8.6","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Ukrainian","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 8.6 percent for Ukrainian.","quote":"Whisper large-v2 8.5 39.3 17.5 99.0 85.8 11.5 8.4 8.6 22.6 90.2 10.3 94.8","where_in_source":"Table 13 (FLEURS), Whisper large-v2 row, column Ukrainian","fact_url":"https://yapapp.cc/facts/whisper-fleurs-ukrainian-8-6","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-19"},{"row":"Whisper large-v2, Malay","value":8.7,"low":null,"high":null,"as_written":"8.7","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Malay","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 8.7 percent for Malay.","quote":"Whisper large-v2 75.6 101.5 28.1 23.1 38.5 16.5 100.7 110.5 38.3 8.7 76.6 115.7 9.5 47.1","where_in_source":"Table 13 (FLEURS), Whisper large-v2 row, column Malay","fact_url":"https://yapapp.cc/facts/whisper-fleurs-malay-8-7","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-19"},{"row":"Edge0/Audio8-ASR-0.1B","value":8.754285714285714,"low":null,"high":null,"as_written":"8.75","unit":"percent","system":"Edge0/Audio8-ASR-0.1B","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: apache-2.0): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, Edge0/Audio8-ASR-0.1B had an average word error rate of 8.75 percent.","quote":"Edge0/Audio8-ASR-0.1B,8.754285714285714,725.77","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-edge0-audio8-asr-0-1b-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"facebook/omniASR-LLM-7B-v2","value":8.78,"low":null,"high":null,"as_written":"8.78","unit":"percent","system":"facebook/omniASR-LLM-7B-v2","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: apache-2.0): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, facebook/omniASR-LLM-7B-v2 had an average word error rate of 8.78 percent.","quote":"facebook/omniASR-LLM-7B-v2,8.78,129.8","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-facebook-omniasr-llm-7b-v2-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, FLEURS Slovak","value":8.82,"low":null,"high":null,"as_written":"8.82","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"FLEURS","language":"Slovak","condition":"Clean read speech","year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 8.82 percent on the FLEURS Slovak test.","quote":"| Language | Fleurs | MLS | CoVoST | ... | **sk** | 8.82% | - | - |","where_in_source":"Multilingual ASR table, Fleurs column, sk row","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-fleurs-sk-8-82","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Canary 1B v2, CoVoST 2","value":8.85,"low":null,"high":null,"as_written":"8.85","unit":"percent","system":"NVIDIA Canary 1B v2","kind":"Open model","benchmark":"CoVoST 2","language":"European languages (average)","condition":null,"year":2025,"conversion":null,"claim":"NVIDIA Canary 1B v2 had an average word error rate of 8.85 percent on the CoVoST 2 test across its CoVoST-13 languages.","quote":"| **WER ↓**          | Fleurs-25 Langs       | CoVoST-13 Langs      | MLS - 6 Langs      | ... | **`Canary-1b-v2`** | 8.40%                 | 8.85%                | 7.27%              |","where_in_source":"ASR table, CoVoST-13 column","fact_url":"https://yapapp.cc/facts/nvidia-canary-1b-v2-covost-2-average-8-85","source_title":"canary-1b-v2 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/canary-1b-v2/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini, CV French","value":8.92,"low":null,"high":null,"as_written":"8.92","unit":"percent","system":"Voxtral Mini","kind":"Open model","benchmark":"Common Voice","language":"French","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, Voxtral Mini had a word error rate of 8.92 percent on Common Voice French.","quote":"Model ar nl en fr de hi it pt es ... Voxtral Mini 63.98 6.03 10.22 8.92 6.07 12.74 5.91 7.76 4.98","where_in_source":"Table 5 (Common Voice), column fr","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-common-voice-fr-8-92","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Small, CV Hindi","value":9.01,"low":null,"high":null,"as_written":"9.01","unit":"percent","system":"Voxtral Small","kind":"Open model","benchmark":"Common Voice","language":"Hindi","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, Voxtral Small had a word error rate of 9.01 percent on Common Voice Hindi.","quote":"Model ar nl en fr de hi it pt es ... Voxtral Small 61.97 3.98 8.58 6.18 3.74 9.01 3.96 6.43 3.31","where_in_source":"Table 5 (Common Voice), column hi","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-small-common-voice-hi-9-01","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"mistralai/Voxtral-Mini-4B-Realtime-2602","value":9.012857142857142,"low":null,"high":null,"as_written":"9.01","unit":"percent","system":"mistralai/Voxtral-Mini-4B-Realtime-2602","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: apache-2.0): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, mistralai/Voxtral-Mini-4B-Realtime-2602 had an average word error rate of 9.01 percent.","quote":"mistralai/Voxtral-Mini-4B-Realtime-2602,9.012857142857142,103.48","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-mistralai-voxtral-mini-4b-realtime-2602-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"Zipformer XL, CTC greedy search","value":9.022857142857143,"low":null,"high":null,"as_written":"9.02","unit":"percent","system":"soundsgoodai/Zipformer-cr-ctc-transducer-XL-290M (fast-gpu-asr, ctc_greedy_search)","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: cc-by-nc-4.0): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, soundsgoodai/Zipformer-cr-ctc-transducer-XL-290M (fast-gpu-asr, ctc_greedy_search) had an average word error rate of 9.02 percent.","quote":"ctc_greedy_search)\",9.022857142857143,20886.37","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-soundsgoodai-zipformer-cr-ctc-transducer-xl-290m-fast-gpu-asr-ctc-greedy-search-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v3, MLS Dutch","value":9.19,"low":null,"high":null,"as_written":"9.19","unit":"percent","system":"Whisper large-v3","kind":"Open model","benchmark":"Multilingual LibriSpeech","language":"Dutch","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 6, Whisper large-v3 had a word error rate of 9.19 percent on Multilingual LibriSpeech Dutch.","quote":"Model nl fr de it pt es ... Whisper large-v3 9.19 5.09 5.72 9.78 7.03 3.89","where_in_source":"Table 6 (Multilingual LibriSpeech), column nl","fact_url":"https://yapapp.cc/facts/voxtral-2025-whisper-large-v3-multilingual-librispeech-nl-9-19","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini, FLEURS Hindi","value":9.26,"low":null,"high":null,"as_written":"9.26","unit":"percent","system":"Voxtral Mini","kind":"Open model","benchmark":"FLEURS","language":"Hindi","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, Voxtral Mini had a word error rate of 9.26 percent on FLEURS Hindi.","quote":"Model ar nl en fr de hi it pt es ... Voxtral Mini 25.40 6.27 3.77 4.87 4.40 9.26 2.51 3.76 3.52","where_in_source":"Table 4 (FLEURS), column hi","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-fleurs-hi-9-26","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini Transcribe, MLS Italian","value":9.28,"low":null,"high":null,"as_written":"9.28","unit":"percent","system":"Voxtral Mini Transcribe","kind":"Open model","benchmark":"Multilingual LibriSpeech","language":"Italian","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 6, Voxtral Mini Transcribe had a word error rate of 9.28 percent on Multilingual LibriSpeech Italian.","quote":"Model nl fr de it pt es ... Voxtral Mini Transcribe 9.63 4.14 5.64 9.28 5.17 3.87","where_in_source":"Table 6 (Multilingual LibriSpeech), column it","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-transcribe-multilingual-librispeech-it-9-28","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Canary-Qwen 2.5B, GigaSpeech English","value":9.41,"low":null,"high":null,"as_written":"9.41","unit":"percent","system":"NVIDIA Canary Qwen 2.5B","kind":"Open model","benchmark":"GigaSpeech","language":"English","condition":"Mixed benchmark","year":2025,"conversion":null,"claim":"NVIDIA Canary Qwen 2.5B had a word error rate of 9.41 percent on the GigaSpeech test set in the Open ASR Leaderboard table.","quote":"| **Version** | **Model**     | **RTFx**   | **Mean**   | **AMI**   | **GigaSpeech**   | **LS Clean**   | **LS Other**   | **Earnings22**   | **SPGISpech**   | **Tedlium**   | **Voxpopuli**   | ... | 2.5.0  | Canary-Qwen-2.5B | 418 | 5.63 | 10.18 | 9.41 | 1.60 | 3.10 | 10.42 | 1.90 | 2.72 | 5.66 |","where_in_source":"Open ASR Leaderboard table, GigaSpeech column","fact_url":"https://yapapp.cc/facts/nvidia-canary-qwen-2-5b-gigaspeech-open-asr-9-41","source_title":"canary-qwen-2.5b model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/canary-qwen-2.5b/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Small, MLS Dutch","value":9.43,"low":null,"high":null,"as_written":"9.43","unit":"percent","system":"Voxtral Small","kind":"Open model","benchmark":"Multilingual LibriSpeech","language":"Dutch","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 6, Voxtral Small had a word error rate of 9.43 percent on Multilingual LibriSpeech Dutch.","quote":"Model nl fr de it pt es ... Voxtral Small 9.43 3.73 5.57 8.44 5.85 3.62","where_in_source":"Table 6 (Multilingual LibriSpeech), column nl","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-small-multilingual-librispeech-nl-9-43","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v2, Norwegian","value":9.5,"low":null,"high":null,"as_written":"9.5","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Norwegian","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 9.5 percent for Norwegian.","quote":"Whisper large-v2 75.6 101.5 28.1 23.1 38.5 16.5 100.7 110.5 38.3 8.7 76.6 115.7 9.5 47.1","where_in_source":"Table 13 (FLEURS), Whisper large-v2 row, column Norwegian","fact_url":"https://yapapp.cc/facts/whisper-fleurs-norwegian-9-5","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-19"},{"row":"Voxtral Mini Transcribe, Earnings-21","value":9.52,"low":null,"high":null,"as_written":"9.52","unit":"percent","system":"Voxtral Mini Transcribe","kind":"Open model","benchmark":"Earnings-21 (10-minute segments)","language":"English","condition":"Specialist vocabulary","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, Voxtral Mini Transcribe had a word error rate of 9.52 percent on Earnings-21 (10-minute segments).","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... Voxtral Mini Transcribe 1.57 3.21 10.04 6.78 11.35 10.03 2.04 9.52 12.18","where_in_source":"Table 3 (long-form English), Earnings-21 (10-minute segments) column","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-transcribe-earnings-21-10-minute-segments-9-52","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Small, Earnings-21 English","value":9.55,"low":null,"high":null,"as_written":"9.55","unit":"percent","system":"Voxtral Small","kind":"Open model","benchmark":"Earnings-21 (10-minute segments)","language":"English","condition":"Specialist vocabulary","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, Voxtral Small had a word error rate of 9.55 percent on Earnings-21 (10-minute segments).","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... Voxtral Small 1.53 3.14 10.27 6.62 11.09 9.64 1.89 9.55 12.48","where_in_source":"Table 3 (long-form English), Earnings-21 (10-minute segments) column","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-small-earnings-21-10-minute-segments-9-55","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v3, VoxPopuli English","value":9.58,"low":null,"high":null,"as_written":"9.58","unit":"percent","system":"Whisper large-v3","kind":"Open model","benchmark":"VoxPopuli","language":"English","condition":"Mixed benchmark","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, Whisper large-v3 had a word error rate of 9.58 percent on VoxPopuli.","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... Whisper large-v3 1.84 3.66 11.60 9.58 13.14 10.88 3.15 9.88 13.07","where_in_source":"Table 3 (short-form English), VoxPopuli column","fact_url":"https://yapapp.cc/facts/voxtral-2025-whisper-large-v3-voxpopuli-9-58","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, GigaSpeech English","value":9.59,"low":null,"high":null,"as_written":"9.59","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"GigaSpeech","language":"English","condition":"Mixed benchmark","year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 9.59 percent on the GigaSpeech test set in the Open ASR Leaderboard table.","quote":"| **Model** | **Avg WER** | **AMI** | **Earnings-22** | **GigaSpeech** | **LS test-clean** | **LS test-other** | **SPGI Speech** | **TEDLIUM-v3** | **VoxPopuli** | ... | `parakeet-tdt-0.6b-v3` | 6.34% | 11.31% | 11.42% | 9.59% | 1.93% | 3.59% | 3.97% | 2.75% | 6.14% |","where_in_source":"Open-ASR-Leaderboard table, GigaSpeech column","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-gigaspeech-open-asr-9-59","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini, Earnings-21 English","value":9.62,"low":null,"high":null,"as_written":"9.62","unit":"percent","system":"Voxtral Mini","kind":"Open model","benchmark":"Earnings-21 (10-minute segments)","language":"English","condition":"Specialist vocabulary","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, Voxtral Mini had a word error rate of 9.62 percent on Earnings-21 (10-minute segments).","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... Voxtral Mini 1.86 4.04 10.68 6.85 11.32 10.59 2.19 9.62 11.84","where_in_source":"Table 3 (long-form English), Earnings-21 (10-minute segments) column","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-earnings-21-10-minute-segments-9-62","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini Transcribe, MLS Dutch","value":9.63,"low":null,"high":null,"as_written":"9.63","unit":"percent","system":"Voxtral Mini Transcribe","kind":"Open model","benchmark":"Multilingual LibriSpeech","language":"Dutch","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 6, Voxtral Mini Transcribe had a word error rate of 9.63 percent on Multilingual LibriSpeech Dutch.","quote":"Model nl fr de it pt es ... Voxtral Mini Transcribe 9.63 4.14 5.64 9.28 5.17 3.87","where_in_source":"Table 6 (Multilingual LibriSpeech), column nl","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-transcribe-multilingual-librispeech-nl-9-63","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Small, CHiME-4 English","value":9.64,"low":null,"high":null,"as_written":"9.64","unit":"percent","system":"Voxtral Small","kind":"Open model","benchmark":"CHiME-4","language":"English","condition":"Noisy or distant","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, Voxtral Small had a word error rate of 9.64 percent on CHiME-4.","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... Voxtral Small 1.53 3.14 10.27 6.62 11.09 9.64 1.89 9.55 12.48","where_in_source":"Table 3 (short-form English), CHiME-4 column","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-small-chime-4-9-64","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v2, Finnish","value":9.7,"low":null,"high":null,"as_written":"9.7","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Finnish","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 9.7 percent for Finnish.","quote":"Whisper large-v2 4.5 12.5 4.2 3.0 21.9 32.9 9.7 13.8 8.3 15.4 102.7 88.9 27.1 21.5","where_in_source":"Table 13 (FLEURS), Whisper large-v2 row, column Finnish","fact_url":"https://yapapp.cc/facts/whisper-fleurs-finnish-9-7","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-19"},{"row":"Whisper large-v3, MLS Italian","value":9.78,"low":null,"high":null,"as_written":"9.78","unit":"percent","system":"Whisper large-v3","kind":"Open model","benchmark":"Multilingual LibriSpeech","language":"Italian","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 6, Whisper large-v3 had a word error rate of 9.78 percent on Multilingual LibriSpeech Italian.","quote":"Model nl fr de it pt es ... Whisper large-v3 9.19 5.09 5.72 9.78 7.03 3.89","where_in_source":"Table 6 (Multilingual LibriSpeech), column it","fact_url":"https://yapapp.cc/facts/voxtral-2025-whisper-large-v3-multilingual-librispeech-it-9-78","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Canary-Qwen 2.5B, noise 0 dB","value":9.83,"low":null,"high":null,"as_written":"9.83","unit":"percent","system":"NVIDIA Canary Qwen 2.5B","kind":"Open model","benchmark":"LibriSpeech test-clean with white noise","language":"English","condition":"Noisy or distant","year":2025,"conversion":null,"claim":"NVIDIA Canary Qwen 2.5B had a word error rate of 9.83 percent on LibriSpeech test-clean with added white noise at 0 dB SNR.","quote":"WER on [Librispeech Test Clean](https://www.openslr.org/12) at different SNR (signal to noise ratio) levels of additive white noise | **Version** | **Model** | **SNR 10** | **SNR 5** | **SNR 0** | **SNR -5** | ... | 2.5.0       | Canary-Qwen-2.5B |    2.41%   |   4.08%   |   9.83%   |    30.60%  |","where_in_source":"White-noise table, SNR 0 dB column","fact_url":"https://yapapp.cc/facts/nvidia-canary-qwen-2-5b-white-noise-snr-0-db-9-83","source_title":"canary-qwen-2.5b model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/canary-qwen-2.5b/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Canary 1B Flash, GigaSpeech English","value":9.85,"low":null,"high":null,"as_written":"9.85","unit":"percent","system":"NVIDIA Canary 1B Flash","kind":"Open model","benchmark":"GigaSpeech","language":"English","condition":"Mixed benchmark","year":null,"conversion":null,"claim":"NVIDIA Canary 1B Flash had a word error rate of 9.85 percent on the GigaSpeech test set in the Open ASR Leaderboard table.","quote":"| **Version** | **Model**     | **RTFx**   | **AMI**   | **GigaSpeech**   | **LS Clean**   | **LS Other**   | **Earnings22**   | **SPGISpech**   | **Tedlium**   | **Voxpopuli**   | ... | nemo-main  | canary-1b-flash | 1045.75 | 13.11 | 9.85 | 1.48 | 2.87 | 12.79 | 1.95 | 3.12 | 5.63 |","where_in_source":"Open ASR Leaderboard table, GigaSpeech column","fact_url":"https://yapapp.cc/facts/nvidia-canary-1b-flash-gigaspeech-open-asr-9-85","source_title":"canary-1b-flash model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"n.d.","source_url":"https://huggingface.co/nvidia/canary-1b-flash/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Vosk small en-us-0.15, 40 MB","value":9.85,"low":null,"high":null,"as_written":"9.85","unit":"percent","system":"Vosk vosk-model-small-en-us-0.15","kind":"Open model","benchmark":"LibriSpeech test-clean","language":"English","condition":"Clean read speech","year":2026,"conversion":null,"claim":"Vosk's small English model vosk-model-small-en-us-0.15, at 40 MB, had a word error rate of 9.85 percent on LibriSpeech test-clean, and it is listed under the Apache 2.0 license.","quote":"vosk-model-small-en-us-0.15 40M 9.85 (librispeech test-clean) 10.38 (tedlium) Lightweight wideband model for Android and RPi Apache 2.0","where_in_source":"Model list table","fact_url":"https://yapapp.cc/facts/vosk-small-en-us-0-15-wer-9-85","source_title":"Vosk models (Alpha Cephei)","source_publisher":"Alpha Cephei","source_year":"2026","source_url":"https://alphacephei.com/vosk/models","source_doi":null,"checked":"2026-10-17"},{"row":"Canary 1B v2, This American Life","value":9.87,"low":null,"high":null,"as_written":"9.87","unit":"percent","system":"NVIDIA Canary 1B v2","kind":"Open model","benchmark":"This American Life","language":"English","condition":"Mixed benchmark","year":2025,"conversion":null,"claim":"NVIDIA Canary 1B v2 had a word error rate of 9.87 percent on the This American Life long-form recordings.","quote":"| **Dataset**             | **WER ↓** | ... | This American Life      | 9.87%     |","where_in_source":"Long-form inference table, This American Life row","fact_url":"https://yapapp.cc/facts/nvidia-canary-1b-v2-this-american-life-long-form-9-87","source_title":"canary-1b-v2 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/canary-1b-v2/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v3, Earnings-21 English","value":9.88,"low":null,"high":null,"as_written":"9.88","unit":"percent","system":"Whisper large-v3","kind":"Open model","benchmark":"Earnings-21 (10-minute segments)","language":"English","condition":"Specialist vocabulary","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, Whisper large-v3 had a word error rate of 9.88 percent on Earnings-21 (10-minute segments).","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... Whisper large-v3 1.84 3.66 11.60 9.58 13.14 10.88 3.15 9.88 13.07","where_in_source":"Table 3 (long-form English), Earnings-21 (10-minute segments) column","fact_url":"https://yapapp.cc/facts/voxtral-2025-whisper-large-v3-earnings-21-10-minute-segments-9-88","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"abr-ai/niagara-84m-batch.en","value":9.881428571428572,"low":null,"high":null,"as_written":"9.88","unit":"percent","system":"abr-ai/niagara-84m-batch.en","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: applied-brain-research-open-license): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, abr-ai/niagara-84m-batch.en had an average word error rate of 9.88 percent.","quote":"abr-ai/niagara-84m-batch.en,9.881428571428572,1940.83","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-abr-ai-niagara-84m-batch-en-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini Transcribe, CHiME-4","value":10.03,"low":null,"high":null,"as_written":"10.03","unit":"percent","system":"Voxtral Mini Transcribe","kind":"Open model","benchmark":"CHiME-4","language":"English","condition":"Noisy or distant","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, Voxtral Mini Transcribe had a word error rate of 10.03 percent on CHiME-4.","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... Voxtral Mini Transcribe 1.57 3.21 10.04 6.78 11.35 10.03 2.04 9.52 12.18","where_in_source":"Table 3 (short-form English), CHiME-4 column","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-transcribe-chime-4-10-03","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini Transcribe, GigaSpeech","value":10.04,"low":null,"high":null,"as_written":"10.04","unit":"percent","system":"Voxtral Mini Transcribe","kind":"Open model","benchmark":"GigaSpeech","language":"English","condition":"Mixed benchmark","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, Voxtral Mini Transcribe had a word error rate of 10.04 percent on GigaSpeech.","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... Voxtral Mini Transcribe 1.57 3.21 10.04 6.78 11.35 10.03 2.04 9.52 12.18","where_in_source":"Table 3 (short-form English), GigaSpeech column","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-transcribe-gigaspeech-10-04","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, MLS Italian","value":10.08,"low":null,"high":null,"as_written":"10.08","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"Multilingual LibriSpeech","language":"Italian","condition":"Clean read speech","year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 10.08 percent on the Multilingual LibriSpeech Italian test.","quote":"| Language | Fleurs | MLS | CoVoST | ... | **it** | 3.00% | 10.08% | 3.69% |","where_in_source":"Multilingual ASR table, MLS column, it row","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-multilingual-librispeech-it-10-08","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini, MLS Dutch","value":10.09,"low":null,"high":null,"as_written":"10.09","unit":"percent","system":"Voxtral Mini","kind":"Open model","benchmark":"Multilingual LibriSpeech","language":"Dutch","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 6, Voxtral Mini had a word error rate of 10.09 percent on Multilingual LibriSpeech Dutch.","quote":"Model nl fr de it pt es ... Voxtral Mini 10.09 5.28 7.09 11.30 6.72 5.12","where_in_source":"Table 6 (Multilingual LibriSpeech), column nl","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-multilingual-librispeech-nl-10-09","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Canary-Qwen 2.5B, AMI English","value":10.18,"low":null,"high":null,"as_written":"10.18","unit":"percent","system":"NVIDIA Canary Qwen 2.5B","kind":"Open model","benchmark":"AMI meetings","language":"English","condition":"Conversational","year":2025,"conversion":null,"claim":"NVIDIA Canary Qwen 2.5B had a word error rate of 10.18 percent on the AMI meetings test set in the Open ASR Leaderboard table.","quote":"| **Version** | **Model**     | **RTFx**   | **Mean**   | **AMI**   | **GigaSpeech**   | **LS Clean**   | **LS Other**   | **Earnings22**   | **SPGISpech**   | **Tedlium**   | **Voxpopuli**   | ... | 2.5.0  | Canary-Qwen-2.5B | 418 | 5.63 | 10.18 | 9.41 | 1.60 | 3.10 | 10.42 | 1.90 | 2.72 | 5.66 |","where_in_source":"Open ASR Leaderboard table, AMI meetings column","fact_url":"https://yapapp.cc/facts/nvidia-canary-qwen-2-5b-ami-meetings-open-asr-10-18","source_title":"canary-qwen-2.5b model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/canary-qwen-2.5b/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini, CV English","value":10.22,"low":null,"high":null,"as_written":"10.22","unit":"percent","system":"Voxtral Mini","kind":"Open model","benchmark":"Common Voice","language":"English","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, Voxtral Mini had a word error rate of 10.22 percent on Common Voice English.","quote":"Model ar nl en fr de hi it pt es ... Voxtral Mini 63.98 6.03 10.22 8.92 6.07 12.74 5.91 7.76 4.98","where_in_source":"Table 5 (Common Voice), column en","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-common-voice-en-10-22","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Small, GigaSpeech English","value":10.27,"low":null,"high":null,"as_written":"10.27","unit":"percent","system":"Voxtral Small","kind":"Open model","benchmark":"GigaSpeech","language":"English","condition":"Mixed benchmark","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, Voxtral Small had a word error rate of 10.27 percent on GigaSpeech.","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... Voxtral Small 1.53 3.14 10.27 6.62 11.09 9.64 1.89 9.55 12.48","where_in_source":"Table 3 (short-form English), GigaSpeech column","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-small-gigaspeech-10-27","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v2, Vietnamese","value":10.3,"low":null,"high":null,"as_written":"10.3","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Vietnamese","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 10.3 percent for Vietnamese.","quote":"Whisper large-v2 8.5 39.3 17.5 99.0 85.8 11.5 8.4 8.6 22.6 90.2 10.3 94.8","where_in_source":"Table 13 (FLEURS), Whisper large-v2 row, column Vietnamese","fact_url":"https://yapapp.cc/facts/whisper-fleurs-vietnamese-10-3","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-19"},{"row":"Voxtral Mini Transcribe, FLEURS Hindi","value":10.32,"low":null,"high":null,"as_written":"10.32","unit":"percent","system":"Voxtral Mini Transcribe","kind":"Open model","benchmark":"FLEURS","language":"Hindi","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, Voxtral Mini Transcribe had a word error rate of 10.32 percent on FLEURS Hindi.","quote":"Model ar nl en fr de hi it pt es ... Voxtral Mini Transcribe 14.64 4.89 3.61 4.22 3.54 10.32 2.31 3.57 2.75","where_in_source":"Table 4 (FLEURS), column hi","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-transcribe-fleurs-hi-10-32","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Canary-Qwen 2.5B, Earnings-22 English","value":10.42,"low":null,"high":null,"as_written":"10.42","unit":"percent","system":"NVIDIA Canary Qwen 2.5B","kind":"Open model","benchmark":"Earnings-22","language":"English","condition":"Specialist vocabulary","year":2025,"conversion":null,"claim":"NVIDIA Canary Qwen 2.5B had a word error rate of 10.42 percent on the Earnings-22 test set in the Open ASR Leaderboard table.","quote":"| **Version** | **Model**     | **RTFx**   | **Mean**   | **AMI**   | **GigaSpeech**   | **LS Clean**   | **LS Other**   | **Earnings22**   | **SPGISpech**   | **Tedlium**   | **Voxpopuli**   | ... | 2.5.0  | Canary-Qwen-2.5B | 418 | 5.63 | 10.18 | 9.41 | 1.60 | 3.10 | 10.42 | 1.90 | 2.72 | 5.66 |","where_in_source":"Open ASR Leaderboard table, Earnings-22 column","fact_url":"https://yapapp.cc/facts/nvidia-canary-qwen-2-5b-earnings-22-open-asr-10-42","source_title":"canary-qwen-2.5b model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/canary-qwen-2.5b/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini Transcribe, CV Hindi","value":10.42,"low":null,"high":null,"as_written":"10.42","unit":"percent","system":"Voxtral Mini Transcribe","kind":"Open model","benchmark":"Common Voice","language":"Hindi","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, Voxtral Mini Transcribe had a word error rate of 10.42 percent on Common Voice Hindi.","quote":"Model ar nl en fr de hi it pt es ... Voxtral Mini Transcribe 62.01 4.71 8.25 7.29 4.85 10.42 4.37 6.70 3.96","where_in_source":"Table 5 (Common Voice), column hi","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-transcribe-common-voice-hi-10-42","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"nvidia/nemotron-3.5-asr-streaming-0.6b","value":10.505714285714287,"low":null,"high":null,"as_written":"10.51","unit":"percent","system":"nvidia/nemotron-3.5-asr-streaming-0.6b","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: OpenMDW-1.1): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, nvidia/nemotron-3.5-asr-streaming-0.6b had an average word error rate of 10.51 percent.","quote":"nvidia/nemotron-3.5-asr-streaming-0.6b,10.505714285714287,1348.36","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-nvidia-nemotron-3-5-asr-streaming-0-6b-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini, CHiME-4 English","value":10.59,"low":null,"high":null,"as_written":"10.59","unit":"percent","system":"Voxtral Mini","kind":"Open model","benchmark":"CHiME-4","language":"English","condition":"Noisy or distant","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, Voxtral Mini had a word error rate of 10.59 percent on CHiME-4.","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... Voxtral Mini 1.86 4.04 10.68 6.85 11.32 10.59 2.19 9.62 11.84","where_in_source":"Table 3 (short-form English), CHiME-4 column","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-chime-4-10-59","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini, GigaSpeech English","value":10.68,"low":null,"high":null,"as_written":"10.68","unit":"percent","system":"Voxtral Mini","kind":"Open model","benchmark":"GigaSpeech","language":"English","condition":"Mixed benchmark","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, Voxtral Mini had a word error rate of 10.68 percent on GigaSpeech.","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... Voxtral Mini 1.86 4.04 10.68 6.85 11.32 10.59 2.19 9.62 11.84","where_in_source":"Table 3 (short-form English), GigaSpeech column","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-gigaspeech-10-68","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Canary 1B v2, GigaSpeech English","value":10.82,"low":null,"high":null,"as_written":"10.82","unit":"percent","system":"NVIDIA Canary 1B v2","kind":"Open model","benchmark":"GigaSpeech","language":"English","condition":"Mixed benchmark","year":2025,"conversion":null,"claim":"NVIDIA Canary 1B v2 had a word error rate of 10.82 percent on the GigaSpeech test set in the Open ASR Leaderboard table.","quote":"|  **WER ↓**   | **RTFx**   | **Mean**   | **AMI**   | **GigaSpeech**   | **LS Clean**   | **LS Other**   | **Earnings22**   | **SPGISpech**   | **Tedlium**   | **Voxpopuli**   | ... | `Canary-1b-v2` | 749 | 7.15 | 16.01 | 10.82 | 2.18 | 3.56 | 11.79 | 2.28 | 4.29 | 6.25 |","where_in_source":"Open ASR Leaderboard table, GigaSpeech column","fact_url":"https://yapapp.cc/facts/nvidia-canary-1b-v2-gigaspeech-open-asr-10-82","source_title":"canary-1b-v2 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/canary-1b-v2/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v3, CHiME-4 English","value":10.88,"low":null,"high":null,"as_written":"10.88","unit":"percent","system":"Whisper large-v3","kind":"Open model","benchmark":"CHiME-4","language":"English","condition":"Noisy or distant","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, Whisper large-v3 had a word error rate of 10.88 percent on CHiME-4.","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... Whisper large-v3 1.84 3.66 11.60 9.58 13.14 10.88 3.15 9.88 13.07","where_in_source":"Table 3 (short-form English), CHiME-4 column","fact_url":"https://yapapp.cc/facts/voxtral-2025-whisper-large-v3-chime-4-10-88","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, FLEURS Czech","value":11.01,"low":null,"high":null,"as_written":"11.01","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"FLEURS","language":"Czech","condition":"Clean read speech","year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 11.01 percent on the FLEURS Czech test.","quote":"| Language | Fleurs | MLS | CoVoST | ... | **cs** | 11.01% | - | - |","where_in_source":"Multilingual ASR table, Fleurs column, cs row","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-fleurs-cs-11-01","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Small, Switchboard English","value":11.09,"low":null,"high":null,"as_written":"11.09","unit":"percent","system":"Voxtral Small","kind":"Open model","benchmark":"Switchboard","language":"English","condition":"Conversational","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, Voxtral Small had a word error rate of 11.09 percent on Switchboard.","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... Voxtral Small 1.53 3.14 10.27 6.62 11.09 9.64 1.89 9.55 12.48","where_in_source":"Table 3 (short-form English), Switchboard column","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-small-switchboard-11-09","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini, MLS Italian","value":11.3,"low":null,"high":null,"as_written":"11.3","unit":"percent","system":"Voxtral Mini","kind":"Open model","benchmark":"Multilingual LibriSpeech","language":"Italian","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 6, Voxtral Mini had a word error rate of 11.30 percent on Multilingual LibriSpeech Italian.","quote":"Model nl fr de it pt es ... Voxtral Mini 10.09 5.28 7.09 11.30 6.72 5.12","where_in_source":"Table 6 (Multilingual LibriSpeech), column it","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-multilingual-librispeech-it-11-30","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, AMI English","value":11.31,"low":null,"high":null,"as_written":"11.31","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"AMI meetings","language":"English","condition":"Conversational","year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 11.31 percent on the AMI meetings test set in the Open ASR Leaderboard table.","quote":"| **Model** | **Avg WER** | **AMI** | **Earnings-22** | **GigaSpeech** | **LS test-clean** | **LS test-other** | **SPGI Speech** | **TEDLIUM-v3** | **VoxPopuli** | ... | `parakeet-tdt-0.6b-v3` | 6.34% | 11.31% | 11.42% | 9.59% | 1.93% | 3.59% | 3.97% | 2.75% | 6.14% |","where_in_source":"Open-ASR-Leaderboard table, AMI meetings column","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-ami-meetings-open-asr-11-31","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini, Switchboard English","value":11.32,"low":null,"high":null,"as_written":"11.32","unit":"percent","system":"Voxtral Mini","kind":"Open model","benchmark":"Switchboard","language":"English","condition":"Conversational","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, Voxtral Mini had a word error rate of 11.32 percent on Switchboard.","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... Voxtral Mini 1.86 4.04 10.68 6.85 11.32 10.59 2.19 9.62 11.84","where_in_source":"Table 3 (short-form English), Switchboard column","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-switchboard-11-32","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v3, CV French","value":11.33,"low":null,"high":null,"as_written":"11.33","unit":"percent","system":"Whisper large-v3","kind":"Open model","benchmark":"Common Voice","language":"French","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, Whisper large-v3 had a word error rate of 11.33 percent on Common Voice French.","quote":"Model ar nl en fr de hi it pt es ... Whisper large-v3 50.58 5.83 22.91 11.33 6.25 46.75 6.81 7.17 5.66","where_in_source":"Table 5 (Common Voice), column fr","fact_url":"https://yapapp.cc/facts/voxtral-2025-whisper-large-v3-common-voice-fr-11-33","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v3-turbo, Croatian (our test)","value":11.34,"low":null,"high":null,"as_written":"11.34","unit":"percent","system":"Whisper large-v3-turbo","kind":"Open model","benchmark":"FLEURS","language":"Croatian","condition":"Clean read speech","year":2026,"conversion":null,"claim":"In our October 2026 benchmark on a Mac with an M2 Pro, Whisper large-v3-turbo had a word error rate of 11.34 percent on 100 Croatian recordings from the FLEURS test set.","quote":null,"where_in_source":null,"fact_url":"https://yapapp.cc/facts/m-fleurs-2026-10-whisper-hr-wer","source_title":null,"source_publisher":null,"source_year":null,"source_url":null,"source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini Transcribe, Switchboard","value":11.35,"low":null,"high":null,"as_written":"11.35","unit":"percent","system":"Voxtral Mini Transcribe","kind":"Open model","benchmark":"Switchboard","language":"English","condition":"Conversational","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, Voxtral Mini Transcribe had a word error rate of 11.35 percent on Switchboard.","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... Voxtral Mini Transcribe 1.57 3.21 10.04 6.78 11.35 10.03 2.04 9.52 12.18","where_in_source":"Table 3 (short-form English), Switchboard column","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-transcribe-switchboard-11-35","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"abr-ai/niagara-38m-batch.en","value":11.388571428571426,"low":null,"high":null,"as_written":"11.39","unit":"percent","system":"abr-ai/niagara-38m-batch.en","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: applied-brain-research-open-license): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, abr-ai/niagara-38m-batch.en had an average word error rate of 11.39 percent.","quote":"abr-ai/niagara-38m-batch.en,11.388571428571426,4561.35","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-abr-ai-niagara-38m-batch-en-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, Earnings-22 English","value":11.42,"low":null,"high":null,"as_written":"11.42","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"Earnings-22","language":"English","condition":"Specialist vocabulary","year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 11.42 percent on the Earnings-22 test set in the Open ASR Leaderboard table.","quote":"| **Model** | **Avg WER** | **AMI** | **Earnings-22** | **GigaSpeech** | **LS test-clean** | **LS test-other** | **SPGI Speech** | **TEDLIUM-v3** | **VoxPopuli** | ... | `parakeet-tdt-0.6b-v3` | 6.34% | 11.31% | 11.42% | 9.59% | 1.93% | 3.59% | 3.97% | 2.75% | 6.14% |","where_in_source":"Open-ASR-Leaderboard table, Earnings-22 column","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-earnings-22-open-asr-11-42","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v3, GigaSpeech English","value":11.6,"low":null,"high":null,"as_written":"11.6","unit":"percent","system":"Whisper large-v3","kind":"Open model","benchmark":"GigaSpeech","language":"English","condition":"Mixed benchmark","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, Whisper large-v3 had a word error rate of 11.60 percent on GigaSpeech.","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... Whisper large-v3 1.84 3.66 11.60 9.58 13.14 10.88 3.15 9.88 13.07","where_in_source":"Table 3 (short-form English), GigaSpeech column","fact_url":"https://yapapp.cc/facts/voxtral-2025-whisper-large-v3-gigaspeech-11-60","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"google/gemma-4-E2B-it","value":11.64625,"low":null,"high":null,"as_written":"11.65","unit":"percent","system":"google/gemma-4-E2B-it","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: apache-2.0): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, google/gemma-4-E2B-it had an average word error rate of 11.65 percent.","quote":"google/gemma-4-E2B-it,11.64625,190.75","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-google-gemma-4-e2b-it-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, noise 0 dB","value":11.66,"low":null,"high":null,"as_written":"11.66","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"Open ASR sets with MUSAN noise (average)","language":"English","condition":"Noisy or distant","year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had an average word error rate of 11.66 percent on the Open ASR sets with MUSAN noise at 0 dB SNR.","quote":"| **SNR Level** | **Avg WER** | **AMI** | **Earnings** | **GigaSpeech** | **LS test-clean** | **LS test-other** | **SPGI** | **Tedlium** | **VoxPopuli** | **Relative Change** | ... | SNR 0 | 11.66% | 24.44% | 17.34% | 13.60% | 4.82% | 10.38% | 8.41% | 5.39% | 8.91% | -83.97% |","where_in_source":"Noise robustness table, SNR 0 row, Avg WER column","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-musan-snr-0-db-11-66","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v2, Slovak","value":11.7,"low":null,"high":null,"as_written":"11.7","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Slovak","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 11.7 percent for Slovak.","quote":"Whisper large-v2 6.7 75.3 102.4 5.4 93.7 4.3 14.4 5.6 156.5 11.7 23.1 121.0 102.9 33.9","where_in_source":"Table 13 (FLEURS), Whisper large-v2 row, column Slovak","fact_url":"https://yapapp.cc/facts/whisper-fleurs-slovak-11-7","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-19"},{"row":"Canary 1B v2, Earnings-22 English","value":11.79,"low":null,"high":null,"as_written":"11.79","unit":"percent","system":"NVIDIA Canary 1B v2","kind":"Open model","benchmark":"Earnings-22","language":"English","condition":"Specialist vocabulary","year":2025,"conversion":null,"claim":"NVIDIA Canary 1B v2 had a word error rate of 11.79 percent on the Earnings-22 test set in the Open ASR Leaderboard table.","quote":"|  **WER ↓**   | **RTFx**   | **Mean**   | **AMI**   | **GigaSpeech**   | **LS Clean**   | **LS Other**   | **Earnings22**   | **SPGISpech**   | **Tedlium**   | **Voxpopuli**   | ... | `Canary-1b-v2` | 749 | 7.15 | 16.01 | 10.82 | 2.18 | 3.56 | 11.79 | 2.28 | 4.29 | 6.25 |","where_in_source":"Open ASR Leaderboard table, Earnings-22 column","fact_url":"https://yapapp.cc/facts/nvidia-canary-1b-v2-earnings-22-open-asr-11-79","source_title":"canary-1b-v2 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/canary-1b-v2/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini, Earnings-22 English","value":11.84,"low":null,"high":null,"as_written":"11.84","unit":"percent","system":"Voxtral Mini","kind":"Open model","benchmark":"Earnings-22 (10-minute segments)","language":"English","condition":"Specialist vocabulary","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, Voxtral Mini had a word error rate of 11.84 percent on Earnings-22 (10-minute segments).","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... Voxtral Mini 1.86 4.04 10.68 6.85 11.32 10.59 2.19 9.62 11.84","where_in_source":"Table 3 (long-form English), Earnings-22 (10-minute segments) column","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-earnings-22-10-minute-segments-11-84","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, FLEURS","value":11.97,"low":null,"high":null,"as_written":"11.97","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"FLEURS","language":"European languages (average)","condition":"Clean read speech","year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had an average word error rate of 11.97 percent on the FLEURS test across its European languages.","quote":"| Language | Fleurs | MLS | CoVoST | ... | **Average WER ↓** | *11.97%* | *7.83%* | *11.98%* |","where_in_source":"Multilingual ASR table, Fleurs column, average row","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-fleurs-average-11-97","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, CoVoST 2","value":11.98,"low":null,"high":null,"as_written":"11.98","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"CoVoST 2","language":"European languages (average)","condition":null,"year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had an average word error rate of 11.98 percent on the CoVoST 2 test across its European languages.","quote":"| Language | Fleurs | MLS | CoVoST | ... | **Average WER ↓** | *11.97%* | *7.83%* | *11.98%* |","where_in_source":"Multilingual ASR table, CoVoST column, average row","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-covost-2-average-11-98","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"facebook/omniASR-CTC-7B-v2","value":12.005714285714285,"low":null,"high":null,"as_written":"12.01","unit":"percent","system":"facebook/omniASR-CTC-7B-v2","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: apache-2.0): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, facebook/omniASR-CTC-7B-v2 had an average word error rate of 12.01 percent.","quote":"facebook/omniASR-CTC-7B-v2,12.005714285714285,522.26","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-facebook-omniasr-ctc-7b-v2-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini Transcribe, Earnings-22","value":12.18,"low":null,"high":null,"as_written":"12.18","unit":"percent","system":"Voxtral Mini Transcribe","kind":"Open model","benchmark":"Earnings-22 (10-minute segments)","language":"English","condition":"Specialist vocabulary","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, Voxtral Mini Transcribe had a word error rate of 12.18 percent on Earnings-22 (10-minute segments).","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... Voxtral Mini Transcribe 1.57 3.21 10.04 6.78 11.35 10.03 2.04 9.52 12.18","where_in_source":"Table 3 (long-form English), Earnings-22 (10-minute segments) column","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-transcribe-earnings-22-10-minute-segments-12-18","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, FLEURS Romanian","value":12.44,"low":null,"high":null,"as_written":"12.44","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"FLEURS","language":"Romanian","condition":"Clean read speech","year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 12.44 percent on the FLEURS Romanian test.","quote":"| Language | Fleurs | MLS | CoVoST | ... | **ro** | 12.44% | - | - |","where_in_source":"Multilingual ASR table, Fleurs column, ro row","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-fleurs-ro-12-44","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, FLEURS Croatian","value":12.46,"low":null,"high":null,"as_written":"12.46","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"FLEURS","language":"Croatian","condition":"Clean read speech","year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 12.46 percent on the FLEURS Croatian test.","quote":"| Language | Fleurs | MLS | CoVoST | ... | **hr** | 12.46% | - | - |","where_in_source":"Multilingual ASR table, Fleurs column, hr row","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-fleurs-hr-12-46","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Small, Earnings-22 English","value":12.48,"low":null,"high":null,"as_written":"12.48","unit":"percent","system":"Voxtral Small","kind":"Open model","benchmark":"Earnings-22 (10-minute segments)","language":"English","condition":"Specialist vocabulary","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, Voxtral Small had a word error rate of 12.48 percent on Earnings-22 (10-minute segments).","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... Voxtral Small 1.53 3.14 10.27 6.62 11.09 9.64 1.89 9.55 12.48","where_in_source":"Table 3 (long-form English), Earnings-22 (10-minute segments) column","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-small-earnings-22-10-minute-segments-12-48","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v2, Greek","value":12.5,"low":null,"high":null,"as_written":"12.5","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Greek","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 12.5 percent for Greek.","quote":"Whisper large-v2 4.5 12.5 4.2 3.0 21.9 32.9 9.7 13.8 8.3 15.4 102.7 88.9 27.1 21.5","where_in_source":"Table 13 (FLEURS), Whisper large-v2 row, column Greek","fact_url":"https://yapapp.cc/facts/whisper-fleurs-greek-12-5","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-19"},{"row":"Parakeet TDT v3, FLEURS Bulgarian","value":12.64,"low":null,"high":null,"as_written":"12.64","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"FLEURS","language":"Bulgarian","condition":"Clean read speech","year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 12.64 percent on the FLEURS Bulgarian test.","quote":"| Language | Fleurs | MLS | CoVoST | ... | **bg** | 12.64% | - | - |","where_in_source":"Multilingual ASR table, Fleurs column, bg row","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-fleurs-bg-12-64","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"google/gemma-4-E4B-it","value":12.66,"low":null,"high":null,"as_written":"12.66","unit":"percent","system":"google/gemma-4-E4B-it","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: apache-2.0): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, google/gemma-4-E4B-it had an average word error rate of 12.66 percent.","quote":"google/gemma-4-E4B-it,12.66,160.69","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-google-gemma-4-e4b-it-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini, CV Hindi","value":12.74,"low":null,"high":null,"as_written":"12.74","unit":"percent","system":"Voxtral Mini","kind":"Open model","benchmark":"Common Voice","language":"Hindi","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, Voxtral Mini had a word error rate of 12.74 percent on Common Voice Hindi.","quote":"Model ar nl en fr de hi it pt es ... Voxtral Mini 63.98 6.03 10.22 8.92 6.07 12.74 5.91 7.76 4.98","where_in_source":"Table 5 (Common Voice), column hi","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-common-voice-hi-12-74","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, MLS Dutch","value":12.78,"low":null,"high":null,"as_written":"12.78","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"Multilingual LibriSpeech","language":"Dutch","condition":"Clean read speech","year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 12.78 percent on the Multilingual LibriSpeech Dutch test.","quote":"| Language | Fleurs | MLS | CoVoST | ... | **nl** | 7.48% | 12.78% | 6.50% |","where_in_source":"Multilingual ASR table, MLS column, nl row","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-multilingual-librispeech-nl-12-78","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Canary 1B Flash, Earnings-22 English","value":12.79,"low":null,"high":null,"as_written":"12.79","unit":"percent","system":"NVIDIA Canary 1B Flash","kind":"Open model","benchmark":"Earnings-22","language":"English","condition":"Specialist vocabulary","year":null,"conversion":null,"claim":"NVIDIA Canary 1B Flash had a word error rate of 12.79 percent on the Earnings-22 test set in the Open ASR Leaderboard table.","quote":"| **Version** | **Model**     | **RTFx**   | **AMI**   | **GigaSpeech**   | **LS Clean**   | **LS Other**   | **Earnings22**   | **SPGISpech**   | **Tedlium**   | **Voxpopuli**   | ... | nemo-main  | canary-1b-flash | 1045.75 | 13.11 | 9.85 | 1.48 | 2.87 | 12.79 | 1.95 | 3.12 | 5.63 |","where_in_source":"Open ASR Leaderboard table, Earnings-22 column","fact_url":"https://yapapp.cc/facts/nvidia-canary-1b-flash-earnings-22-open-asr-12-79","source_title":"canary-1b-flash model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"n.d.","source_url":"https://huggingface.co/nvidia/canary-1b-flash/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v3, Earnings-22 English","value":13.07,"low":null,"high":null,"as_written":"13.07","unit":"percent","system":"Whisper large-v3","kind":"Open model","benchmark":"Earnings-22 (10-minute segments)","language":"English","condition":"Specialist vocabulary","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, Whisper large-v3 had a word error rate of 13.07 percent on Earnings-22 (10-minute segments).","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... Whisper large-v3 1.84 3.66 11.60 9.58 13.14 10.88 3.15 9.88 13.07","where_in_source":"Table 3 (long-form English), Earnings-22 (10-minute segments) column","fact_url":"https://yapapp.cc/facts/voxtral-2025-whisper-large-v3-earnings-22-10-minute-segments-13-07","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Canary 1B Flash, AMI English","value":13.11,"low":null,"high":null,"as_written":"13.11","unit":"percent","system":"NVIDIA Canary 1B Flash","kind":"Open model","benchmark":"AMI meetings","language":"English","condition":"Conversational","year":null,"conversion":null,"claim":"NVIDIA Canary 1B Flash had a word error rate of 13.11 percent on the AMI meetings test set in the Open ASR Leaderboard table.","quote":"| **Version** | **Model**     | **RTFx**   | **AMI**   | **GigaSpeech**   | **LS Clean**   | **LS Other**   | **Earnings22**   | **SPGISpech**   | **Tedlium**   | **Voxpopuli**   | ... | nemo-main  | canary-1b-flash | 1045.75 | 13.11 | 9.85 | 1.48 | 2.87 | 12.79 | 1.95 | 3.12 | 5.63 |","where_in_source":"Open ASR Leaderboard table, AMI meetings column","fact_url":"https://yapapp.cc/facts/nvidia-canary-1b-flash-ami-meetings-open-asr-13-11","source_title":"canary-1b-flash model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"n.d.","source_url":"https://huggingface.co/nvidia/canary-1b-flash/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v3, Switchboard English","value":13.14,"low":null,"high":null,"as_written":"13.14","unit":"percent","system":"Whisper large-v3","kind":"Open model","benchmark":"Switchboard","language":"English","condition":"Conversational","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, Whisper large-v3 had a word error rate of 13.14 percent on Switchboard.","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... Whisper large-v3 1.84 3.66 11.60 9.58 13.14 10.88 3.15 9.88 13.07","where_in_source":"Table 3 (short-form English), Switchboard column","fact_url":"https://yapapp.cc/facts/voxtral-2025-whisper-large-v3-switchboard-13-14","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, FLEURS Finnish","value":13.21,"low":null,"high":null,"as_written":"13.21","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"FLEURS","language":"Finnish","condition":"Clean read speech","year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 13.21 percent on the FLEURS Finnish test.","quote":"| Language | Fleurs | MLS | CoVoST | ... | **fi** | 13.21% | - | - |","where_in_source":"Multilingual ASR table, Fleurs column, fi row","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-fleurs-fi-13-21","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"abr-ai/niagara-19m-batch.en","value":13.27714285714286,"low":null,"high":null,"as_written":"13.28","unit":"percent","system":"abr-ai/niagara-19m-batch.en","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: applied-brain-research-open-license): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, abr-ai/niagara-19m-batch.en had an average word error rate of 13.28 percent.","quote":"abr-ai/niagara-19m-batch.en,13.27714285714286,4348.44","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-abr-ai-niagara-19m-batch-en-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v2, Czech","value":13.3,"low":null,"high":null,"as_written":"13.3","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Czech","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 13.3 percent for Czech.","quote":"Whisper large-v2 36.7 140.3 16.0 106.2 23.4 45.4 14.6 104.1 15.7 7.3 14.7 13.3 33.0 13.8","where_in_source":"Table 13 (FLEURS), Whisper large-v2 row, column Czech","fact_url":"https://yapapp.cc/facts/whisper-fleurs-czech-13-3","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-19"},{"row":"Whisper large-v2, Croatian","value":13.4,"low":null,"high":null,"as_written":"13.4","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Croatian","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 13.4 percent for Croatian.","quote":"Whisper large-v2 13.4 17.0 44.6 7.1 38.2 4.0 5.3 nan 105.0 37.7 99.7 37.0 14.3 88.0","where_in_source":"Table 13 (FLEURS), Whisper large-v2 row, column Croatian","fact_url":"https://yapapp.cc/facts/whisper-fleurs-croatian-13-4","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-19"},{"row":"Voxtral Small, FLEURS Arabic","value":13.44,"low":null,"high":null,"as_written":"13.44","unit":"percent","system":"Voxtral Small","kind":"Open model","benchmark":"FLEURS","language":"Arabic","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, Voxtral Small had a word error rate of 13.44 percent on FLEURS Arabic.","quote":"Model ar nl en fr de hi it pt es ... Voxtral Small 13.44 4.94 3.35 4.03 3.38 7.69 2.62 3.79 2.72","where_in_source":"Table 4 (FLEURS), column ar","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-small-fleurs-ar-13-44","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Canary 1B v2, Earnings-22 long-form","value":13.78,"low":null,"high":null,"as_written":"13.78","unit":"percent","system":"NVIDIA Canary 1B v2","kind":"Open model","benchmark":"Earnings-22 long-form","language":"English","condition":"Specialist vocabulary","year":2025,"conversion":null,"claim":"NVIDIA Canary 1B v2 had a word error rate of 13.78 percent on the Earnings-22 long-form long-form recordings.","quote":"| **Dataset**             | **WER ↓** | ... | Earnings-22             | 13.78%    |","where_in_source":"Long-form inference table, Earnings-22 row","fact_url":"https://yapapp.cc/facts/nvidia-canary-1b-v2-earnings-22-long-form-long-form-13-78","source_title":"canary-1b-v2 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/canary-1b-v2/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v2, Danish","value":13.8,"low":null,"high":null,"as_written":"13.8","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Danish","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 13.8 percent for Danish.","quote":"Whisper large-v2 36.7 140.3 16.0 106.2 23.4 45.4 14.6 104.1 15.7 7.3 14.7 13.3 33.0 13.8","where_in_source":"Table 13 (FLEURS), Whisper large-v2 row, column Danish","fact_url":"https://yapapp.cc/facts/whisper-fleurs-danish-13-8","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-19"},{"row":"Whisper large-v2, Switchboard","value":13.8,"low":null,"high":null,"as_written":"13.8","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"Switchboard (Hub5 2000)","language":"English","condition":null,"year":2022,"conversion":"The source's Switchboard test is the Hub5 2000 set, which the same table also uses for CallHome. The language is English because Switchboard is English telephone speech. The figure is the second column of the row.","claim":"On the Switchboard test set, Whisper Large V2 had a word error rate of 13.8 percent.","quote":"Switchboard 28.3 13.8 51.2 CallHome 34.8 17.6 49.4","where_in_source":"Table 2, effective robustness across datasets","fact_url":"https://yapapp.cc/facts/whisper-switchboard-wer-13-8","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision","source_publisher":"arXiv (eess.AS)","source_year":"2022","source_url":"https://arxiv.org/abs/2212.04356","source_doi":null,"checked":"2026-10-16"},{"row":"Whisper large-v2, Korean","value":14.3,"low":null,"high":null,"as_written":"14.3","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Korean","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the Fleurs benchmark, the Whisper large-v2 model had a word error rate of 14.3 percent for Korean.","quote":"Model Croatian Hungarian Armenian Indonesian Icelandic Italian Japanese Javanese Georgian Kazakh Khmer Kannada Korean Luxembourgish ... Whisper large-v2 13.4 17.0 44.6 7.1 38.2 4.0 5.3 nan 105.0 37.7 99.7 37.0 14.3 88.0","where_in_source":"Table 13 (Fleurs), row Whisper large-v2, column Korean","fact_url":"https://yapapp.cc/facts/whisper-fleurs-korean-14-3","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-17"},{"row":"Whisper large-v2, Romanian","value":14.4,"low":null,"high":null,"as_written":"14.4","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Romanian","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 14.4 percent for Romanian.","quote":"Whisper large-v2 6.7 75.3 102.4 5.4 93.7 4.3 14.4 5.6 156.5 11.7 23.1 121.0 102.9 33.9","where_in_source":"Table 13 (FLEURS), Whisper large-v2 row, column Romanian","fact_url":"https://yapapp.cc/facts/whisper-fleurs-romanian-14-4","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-19"},{"row":"Whisper large-v2, Bulgarian","value":14.6,"low":null,"high":null,"as_written":"14.6","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Bulgarian","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 14.6 percent for Bulgarian.","quote":"Whisper large-v2 36.7 140.3 16.0 106.2 23.4 45.4 14.6 104.1 15.7 7.3 14.7 13.3 33.0 13.8","where_in_source":"Table 13 (FLEURS), Whisper large-v2 row, column Bulgarian","fact_url":"https://yapapp.cc/facts/whisper-fleurs-bulgarian-14-6","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-19"},{"row":"Voxtral Mini Transcribe, FLEURS Arabic","value":14.64,"low":null,"high":null,"as_written":"14.64","unit":"percent","system":"Voxtral Mini Transcribe","kind":"Open model","benchmark":"FLEURS","language":"Arabic","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, Voxtral Mini Transcribe had a word error rate of 14.64 percent on FLEURS Arabic.","quote":"Model ar nl en fr de hi it pt es ... Voxtral Mini Transcribe 14.64 4.89 3.61 4.22 3.54 10.32 2.31 3.57 2.75","where_in_source":"Table 4 (FLEURS), column ar","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-transcribe-fleurs-ar-14-64","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, FLEURS Swedish","value":15.08,"low":null,"high":null,"as_written":"15.08","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"FLEURS","language":"Swedish","condition":"Clean read speech","year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 15.08 percent on the FLEURS Swedish test.","quote":"| Language | Fleurs | MLS | CoVoST | ... | **sv** | 15.08% | - | 20.16% |","where_in_source":"Multilingual ASR table, Fleurs column, sv row","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-fleurs-sv-15-08","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v2, Galician","value":15.4,"low":null,"high":null,"as_written":"15.4","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Galician","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 15.4 percent for Galician.","quote":"Whisper large-v2 4.5 12.5 4.2 3.0 21.9 32.9 9.7 13.8 8.3 15.4 102.7 88.9 27.1 21.5","where_in_source":"Table 13 (FLEURS), Whisper large-v2 row, column Galician","fact_url":"https://yapapp.cc/facts/whisper-fleurs-galician-15-4","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-19"},{"row":"Whisper large-v3, FLEURS Arabic","value":15.44,"low":null,"high":null,"as_written":"15.44","unit":"percent","system":"Whisper large-v3","kind":"Open model","benchmark":"FLEURS","language":"Arabic","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's FLEURS table, Whisper large-v3 had a word error rate of 15.44 percent in Arabic.","quote":"Model ar nl en fr de hi it pt es Whisper large-v3 15.44 5.87 4.00 5.55 5.46 28.87 2.71 3.90 2.81 GPT-4o mini Transcribe 14.02 5.54 3.19 4.51 3.76 12.36 2.02 3.54 2.58 Gemini 2.5 Flash 25.25 6.20 4.64 6.17 4.74 6.76","where_in_source":"Table 4 (FLEURS), column ar","fact_url":"https://yapapp.cc/facts/voxtral-2025-whisper-large-v3-arabic-15-44","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-17"},{"row":"Whisper large-v2, Bosnian","value":15.7,"low":null,"high":null,"as_written":"15.7","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Bosnian","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 15.7 percent for Bosnian.","quote":"Whisper large-v2 36.7 140.3 16.0 106.2 23.4 45.4 14.6 104.1 15.7 7.3 14.7 13.3 33.0 13.8","where_in_source":"Table 13 (FLEURS), Whisper large-v2 row, column Bosnian","fact_url":"https://yapapp.cc/facts/whisper-fleurs-bosnian-15-7","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-19"},{"row":"Parakeet TDT v3, FLEURS Hungarian","value":15.72,"low":null,"high":null,"as_written":"15.72","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"FLEURS","language":"Hungarian","condition":"Clean read speech","year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 15.72 percent on the FLEURS Hungarian test.","quote":"| Language | Fleurs | MLS | CoVoST | ... | **hu** | 15.72% | - | - |","where_in_source":"Multilingual ASR table, Fleurs column, hu row","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-fleurs-hu-15-72","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v2, Arabic","value":16,"low":null,"high":null,"as_written":"16","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Arabic","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the Fleurs benchmark, the Whisper large-v2 model had a word error rate of 16.0 percent for Arabic.","quote":"Model Afrikaans Amharic Arabic Assamese Azerbaijani Belarusian Bulgarian Bengali Bosnian Catalan Chinese Czech Welsh Danish ... Whisper large-v2 36.7 140.3 16.0 106.2 23.4 45.4 14.6 104.1 15.7 7.3 14.7 13.3 33.0 13.8","where_in_source":"Table 13 (Fleurs)","fact_url":"https://yapapp.cc/facts/whisper-fleurs-arabic-16-0","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-16"},{"row":"Canary 1B v2, AMI English","value":16.01,"low":null,"high":null,"as_written":"16.01","unit":"percent","system":"NVIDIA Canary 1B v2","kind":"Open model","benchmark":"AMI meetings","language":"English","condition":"Conversational","year":2025,"conversion":null,"claim":"NVIDIA Canary 1B v2 had a word error rate of 16.01 percent on the AMI meetings test set in the Open ASR Leaderboard table.","quote":"|  **WER ↓**   | **RTFx**   | **Mean**   | **AMI**   | **GigaSpeech**   | **LS Clean**   | **LS Other**   | **Earnings22**   | **SPGISpech**   | **Tedlium**   | **Voxpopuli**   | ... | `Canary-1b-v2` | 749 | 7.15 | 16.01 | 10.82 | 2.18 | 3.56 | 11.79 | 2.28 | 4.29 | 6.25 |","where_in_source":"Open ASR Leaderboard table, AMI meetings column","fact_url":"https://yapapp.cc/facts/nvidia-canary-1b-v2-ami-meetings-open-asr-16-01","source_title":"canary-1b-v2 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/canary-1b-v2/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v2, Macedonian","value":16.5,"low":null,"high":null,"as_written":"16.5","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Macedonian","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 16.5 percent for Macedonian.","quote":"Whisper large-v2 75.6 101.5 28.1 23.1 38.5 16.5 100.7 110.5 38.3 8.7 76.6 115.7 9.5 47.1","where_in_source":"Table 13 (FLEURS), Whisper large-v2 row, column Macedonian","fact_url":"https://yapapp.cc/facts/whisper-fleurs-macedonian-16-5","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-19"},{"row":"abr-ai/niagara-9m-batch.en","value":16.634285714285713,"low":null,"high":null,"as_written":"16.63","unit":"percent","system":"abr-ai/niagara-9m-batch.en","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: applied-brain-research-open-license): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, abr-ai/niagara-9m-batch.en had an average word error rate of 16.63 percent.","quote":"abr-ai/niagara-9m-batch.en,16.634285714285713,7217.02","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-abr-ai-niagara-9m-batch-en-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v2, AMI IHM","value":16.9,"low":null,"high":null,"as_written":"16.9","unit":"percent","system":"Whisper Large V2","kind":"Open model","benchmark":"AMI meetings (IHM)","language":"English","condition":"Conversational","year":2022,"conversion":"IHM is the individual headset microphone condition. Language is taken from the test set name: AMI is an English meeting corpus.","claim":"On the AMI meeting corpus recorded with individual headset microphones (IHM), Whisper Large V2 had a word error rate of 16.9 percent.","quote":"CORAAL 35.6 16.2 54.5 AMI IHM 37.0 16.9 54.3 Switchboard 28.3 13.8 51.2","where_in_source":"Table 2, effective robustness across datasets, AMI IHM row","fact_url":"https://yapapp.cc/facts/whisper-ami-ihm-16-9","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision","source_publisher":"arXiv (eess.AS)","source_year":"2022","source_url":"https://arxiv.org/abs/2212.04356","source_doi":null,"checked":"2026-10-17"},{"row":"Whisper large-v2, Hungarian","value":17,"low":null,"high":null,"as_written":"17","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Hungarian","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 17.0 percent for Hungarian.","quote":"Whisper large-v2 13.4 17.0 44.6 7.1 38.2 4.0 5.3 nan 105.0 37.7 99.7 37.0 14.3 88.0","where_in_source":"Table 13 (FLEURS), Whisper large-v2 row, column Hungarian","fact_url":"https://yapapp.cc/facts/whisper-fleurs-hungarian-17-0","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-19"},{"row":"Whisper large-v2, Tamil","value":17.5,"low":null,"high":null,"as_written":"17.5","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Tamil","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the Fleurs benchmark, the Whisper large-v2 model had a word error rate of 17.5 percent for Tamil.","quote":"Model Swedish Swahili Tamil Telugu Tajik Thai Turkish Ukrainian Urdu Uzbek Vietnamese Yoruba ... Whisper large-v2 8.5 39.3 17.5 99.0 85.8 11.5 8.4 8.6 22.6 90.2 10.3 94.8","where_in_source":"Table 13 (Fleurs)","fact_url":"https://yapapp.cc/facts/whisper-fleurs-tamil-17-5","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-16"},{"row":"Whisper large-v2, CallHome","value":17.6,"low":null,"high":null,"as_written":"17.6","unit":"percent","system":"Whisper Large V2","kind":"Open model","benchmark":"CallHome","language":"English","condition":"Conversational","year":2022,"conversion":"Language is taken from the paper's English evaluation; the fact does not name the language.","claim":"On the CallHome test set of conversations between friends and family, Whisper Large V2 had a word error rate of 17.6 percent.","quote":"Switchboard 28.3 13.8 51.2 CallHome 34.8 17.6 49.4","where_in_source":"Table 2, effective robustness across datasets","fact_url":"https://yapapp.cc/facts/whisper-2022-callhome-wer-17-6","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-17"},{"row":"Parakeet TDT v3, FLEURS Estonian","value":17.73,"low":null,"high":null,"as_written":"17.73","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"FLEURS","language":"Estonian","condition":"Clean read speech","year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 17.73 percent on the FLEURS Estonian test.","quote":"| Language | Fleurs | MLS | CoVoST | ... | **et** | 17.73% | - | 22.04% |","where_in_source":"Multilingual ASR table, Fleurs column, et row","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-fleurs-et-17-73","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, FLEURS Danish","value":18.41,"low":null,"high":null,"as_written":"18.41","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"FLEURS","language":"Danish","condition":"Clean read speech","year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 18.41 percent on the FLEURS Danish test.","quote":"| Language | Fleurs | MLS | CoVoST | ... | **da** | 18.41% | - | - |","where_in_source":"Multilingual ASR table, Fleurs column, da row","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-fleurs-da-18-41","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v3-turbo, Latvian (our test)","value":18.64,"low":null,"high":null,"as_written":"18.64","unit":"percent","system":"Whisper large-v3-turbo","kind":"Open model","benchmark":"FLEURS","language":"Latvian","condition":"Clean read speech","year":2026,"conversion":null,"claim":"In our October 2026 benchmark on a Mac with an M2 Pro, Whisper large-v3-turbo had a word error rate of 18.64 percent on 100 Latvian recordings from the FLEURS test set.","quote":null,"where_in_source":null,"fact_url":"https://yapapp.cc/facts/m-fleurs-2026-10-whisper-lv-wer","source_title":null,"source_publisher":null,"source_year":null,"source_url":null,"source_doi":null,"checked":"2026-10-19"},{"row":"Whisper, whispered speech","value":18.8,"low":null,"high":null,"as_written":"18.8","unit":"percent","system":"Whisper (baseline)","kind":"Open model","benchmark":"wTIMIT and CHAINS whispered speech","language":"English","condition":null,"year":2024,"conversion":null,"claim":"In a 2024 baseline test, the OpenAI Whisper model had a word error rate of 18.8 percent on whispered speech.","quote":"achieving a Word Error Rate (WER) of 18.8% and a Character Error Rate (CER) of 4.24% on whispered speech","where_in_source":"Abstract","fact_url":"https://yapapp.cc/facts/farhadipour-2024-whispered-wer-18-8","source_title":"Leveraging Self-Supervised Models for Automatic Whispered Speech Recognition","source_publisher":"arXiv (Electrical Engineering and Systems Science, Audio and Speech Processing)","source_year":"2024","source_url":"https://arxiv.org/abs/2407.21211","source_doi":null,"checked":"2026-10-16"},{"row":"Canary 1B v2, noise -5 dB","value":19.38,"low":null,"high":null,"as_written":"19.38","unit":"percent","system":"NVIDIA Canary 1B v2","kind":"Open model","benchmark":"LibriSpeech test-clean with MUSAN noise","language":"English","condition":"Noisy or distant","year":2025,"conversion":null,"claim":"NVIDIA Canary 1B v2 had a word error rate of 19.38 percent on LibriSpeech Clean with MUSAN noise at -5 dB SNR.","quote":"Performance across different Signal-to-Noise Ratios (SNR) using MUSAN music and noise samples \\[16] on the [LibriSpeech Clean test set](https://www.openslr.org/12). ... | **SNR (dB)**       | 100      | 10    | 5     | 0     | -5     | ... | **`Canary-1b-v2`** | 2.18% | 2.29% | 2.80% | 5.08% | 19.38% |","where_in_source":"Noise robustness table, SNR -5 dB column","fact_url":"https://yapapp.cc/facts/nvidia-canary-1b-v2-musan-snr-minus-5-db-19-38","source_title":"canary-1b-v2 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/canary-1b-v2/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, Slovenian (our test)","value":19.83,"low":null,"high":null,"as_written":"19.83","unit":"percent","system":"Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"FLEURS","language":"Slovenian","condition":"Clean read speech","year":2026,"conversion":null,"claim":"In our October 2026 benchmark on a Mac with an M2 Pro, Parakeet TDT 0.6B v3 had a word error rate of 19.83 percent on 100 Slovenian recordings from the FLEURS test set.","quote":null,"where_in_source":null,"fact_url":"https://yapapp.cc/facts/m-fleurs-2026-10-parakeet-sl-wer","source_title":null,"source_publisher":null,"source_year":null,"source_url":null,"source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, noise -5 dB","value":19.88,"low":null,"high":null,"as_written":"19.88","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"Open ASR sets with MUSAN noise (average)","language":"English","condition":"Noisy or distant","year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had an average word error rate of 19.88 percent on the Open ASR sets with MUSAN noise at -5 dB SNR.","quote":"| **SNR Level** | **Avg WER** | **AMI** | **Earnings** | **GigaSpeech** | **LS test-clean** | **LS test-other** | **SPGI** | **Tedlium** | **VoxPopuli** | **Relative Change** | ... | SNR -5 | 19.88% | 34.91% | 26.92% | 21.41% | 12.21% | 19.98% | 16.96% | 11.36% | 15.30% | -213.64% |","where_in_source":"Noise robustness table, SNR -5 row, Avg WER column","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-musan-snr-minus-5-db-19-88","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, CoVoST 2 Swedish","value":20.16,"low":null,"high":null,"as_written":"20.16","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"CoVoST 2","language":"Swedish","condition":null,"year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 20.16 percent on the CoVoST 2 Swedish test.","quote":"| Language | Fleurs | MLS | CoVoST | ... | **sv** | 15.08% | - | 20.16% |","where_in_source":"Multilingual ASR table, CoVoST column, sv row","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-covost-2-sv-20-16","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, FLEURS Lithuanian","value":20.35,"low":null,"high":null,"as_written":"20.35","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"FLEURS","language":"Lithuanian","condition":"Clean read speech","year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 20.35 percent on the FLEURS Lithuanian test.","quote":"| Language | Fleurs | MLS | CoVoST | ... | **lt** | 20.35% | - | - |","where_in_source":"Multilingual ASR table, Fleurs column, lt row","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-fleurs-lt-20-35","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, FLEURS Maltese","value":20.46,"low":null,"high":null,"as_written":"20.46","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"FLEURS","language":"Maltese","condition":"Clean read speech","year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 20.46 percent on the FLEURS Maltese test.","quote":"| Language | Fleurs | MLS | CoVoST | ... | **mt** | 20.46% | - | - |","where_in_source":"Multilingual ASR table, Fleurs column, mt row","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-fleurs-mt-20-46","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, FLEURS Greek","value":20.7,"low":null,"high":null,"as_written":"20.7","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"FLEURS","language":"Greek","condition":"Clean read speech","year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 20.70 percent on the FLEURS Greek test.","quote":"| Language | Fleurs | MLS | CoVoST | ... | **el** | 20.70% | - | - |","where_in_source":"Multilingual ASR table, Fleurs column, el row","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-fleurs-el-20-70","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"google/gemma-4-12B-it","value":20.86875,"low":null,"high":null,"as_written":"20.87","unit":"percent","system":"google/gemma-4-12B-it","kind":"Open model","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: apache-2.0): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, google/gemma-4-12B-it had an average word error rate of 20.87 percent.","quote":"google/gemma-4-12B-it,20.86875,106.08","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-google-gemma-4-12b-it-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v2, Hindi","value":21.5,"low":null,"high":null,"as_written":"21.5","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Hindi","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the Fleurs benchmark, the Whisper large-v2 model had a word error rate of 21.5 percent for Hindi.","quote":"Model German Greek English Spanish Estonian Persian Finnish Tagalog French Galician Gujarati Hausa Hebrew Hindi ... Whisper large-v2 4.5 12.5 4.2 3.0 21.9 32.9 9.7 13.8 8.3 15.4 102.7 88.9 27.1 21.5","where_in_source":"Table 13 (Fleurs)","fact_url":"https://yapapp.cc/facts/whisper-fleurs-hindi-21-5","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-16"},{"row":"Whisper large-v2, Estonian","value":21.9,"low":null,"high":null,"as_written":"21.9","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Estonian","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 21.9 percent for Estonian.","quote":"Whisper large-v2 4.5 12.5 4.2 3.0 21.9 32.9 9.7 13.8 8.3 15.4 102.7 88.9 27.1 21.5","where_in_source":"Table 13 (FLEURS), Whisper large-v2 row, column Estonian","fact_url":"https://yapapp.cc/facts/whisper-fleurs-estonian-21-9","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-19"},{"row":"Parakeet TDT v3, CoVoST 2 Estonian","value":22.04,"low":null,"high":null,"as_written":"22.04","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"CoVoST 2","language":"Estonian","condition":null,"year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 22.04 percent on the CoVoST 2 Estonian test.","quote":"| Language | Fleurs | MLS | CoVoST | ... | **et** | 17.73% | - | 22.04% |","where_in_source":"Multilingual ASR table, CoVoST column, et row","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-covost-2-et-22-04","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v2, Urdu","value":22.6,"low":null,"high":null,"as_written":"22.6","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Urdu","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the Fleurs benchmark, the Whisper large-v2 model had a word error rate of 22.6 percent for Urdu.","quote":"Model Swedish Swahili Tamil Telugu Tajik Thai Turkish Ukrainian Urdu Uzbek Vietnamese Yoruba ... Whisper large-v2 8.5 39.3 17.5 99.0 85.8 11.5 8.4 8.6 22.6 90.2 10.3 94.8","where_in_source":"Table 13 (Fleurs)","fact_url":"https://yapapp.cc/facts/whisper-fleurs-urdu-22-6","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-16"},{"row":"Parakeet TDT v3, FLEURS Latvian","value":22.84,"low":null,"high":null,"as_written":"22.84","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"FLEURS","language":"Latvian","condition":"Clean read speech","year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 22.84 percent on the FLEURS Latvian test.","quote":"| Language | Fleurs | MLS | CoVoST | ... | **lv** | 22.84% | - | 38.36% |","where_in_source":"Multilingual ASR table, Fleurs column, lv row","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-fleurs-lv-22-84","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v3, CV English","value":22.91,"low":null,"high":null,"as_written":"22.91","unit":"percent","system":"Whisper large-v3","kind":"Open model","benchmark":"Common Voice","language":"English","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, Whisper large-v3 had a word error rate of 22.91 percent on Common Voice English.","quote":"Model ar nl en fr de hi it pt es ... Whisper large-v3 50.58 5.83 22.91 11.33 6.25 46.75 6.81 7.17 5.66","where_in_source":"Table 5 (Common Voice), column en","fact_url":"https://yapapp.cc/facts/voxtral-2025-whisper-large-v3-common-voice-en-22-91","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v3, zero-shot","value":23,"low":null,"high":null,"as_written":"23","unit":"percent","system":"Whisper large-v3 (zero-shot)","kind":"Open model","benchmark":"STT4SG-350 test set","language":"Swiss German","condition":"Accented or dialect","year":2024,"conversion":"The table gives 0.230 as a fraction, which is 23.0 percent.","claim":"In Dolev and colleagues' tests, Whisper large-v3 used without any training on Swiss German had a word error rate of 23.0 percent on the STT4SG-350 test set.","quote":"Whisper large-v3 zero-shot 0.230 63.1 This work","where_in_source":"Table 2, STT4SG-350 row","fact_url":"https://yapapp.cc/facts/dolev-2024-whisper-large-v3-swiss-german-23-percent","source_title":"Does Whisper Understand Swiss German? An Automatic, Qualitative and Human Evaluation","source_publisher":"Proceedings of the 11th Workshop on NLP for Similar Languages, Varieties and Dialects (VarDial 2024)","source_year":"2024","source_url":"https://arxiv.org/html/2404.19310v2","source_doi":null,"checked":"2026-10-17"},{"row":"Whisper large-v2, Latvian","value":23.1,"low":null,"high":null,"as_written":"23.1","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Latvian","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 23.1 percent for Latvian.","quote":"Whisper large-v2 75.6 101.5 28.1 23.1 38.5 16.5 100.7 110.5 38.3 8.7 76.6 115.7 9.5 47.1","where_in_source":"Table 13 (FLEURS), Whisper large-v2 row, column Latvian","fact_url":"https://yapapp.cc/facts/whisper-fleurs-latvian-23-1","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-19"},{"row":"Whisper large-v2, Azerbaijani","value":23.4,"low":null,"high":null,"as_written":"23.4","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Azerbaijani","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 23.4 percent for Azerbaijani.","quote":"Whisper large-v2 36.7 140.3 16.0 106.2 23.4 45.4 14.6 104.1 15.7 7.3 14.7 13.3 33.0 13.8","where_in_source":"Table 13 (FLEURS), Whisper large-v2 row, column Azerbaijani","fact_url":"https://yapapp.cc/facts/whisper-fleurs-azerbaijani-23-4","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-19"},{"row":"Whisper large-v3-turbo, Slovenian (our test)","value":23.51,"low":null,"high":null,"as_written":"23.51","unit":"percent","system":"Whisper large-v3-turbo","kind":"Open model","benchmark":"FLEURS","language":"Slovenian","condition":"Clean read speech","year":2026,"conversion":null,"claim":"In our October 2026 benchmark on a Mac with an M2 Pro, Whisper large-v3-turbo had a word error rate of 23.51 percent on 100 Slovenian recordings from the FLEURS test set.","quote":null,"where_in_source":null,"fact_url":"https://yapapp.cc/facts/m-fleurs-2026-10-whisper-sl-wer","source_title":null,"source_publisher":null,"source_year":null,"source_url":null,"source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT v3, FLEURS Slovenian","value":24.03,"low":null,"high":null,"as_written":"24.03","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"FLEURS","language":"Slovenian","condition":"Clean read speech","year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 24.03 percent on the FLEURS Slovenian test.","quote":"| Language | Fleurs | MLS | CoVoST | ... | **sl** | 24.03% | - | 31.80% |","where_in_source":"Multilingual ASR table, Fleurs column, sl row","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-fleurs-sl-24-03","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini, FLEURS Arabic","value":25.4,"low":null,"high":null,"as_written":"25.4","unit":"percent","system":"Voxtral Mini","kind":"Open model","benchmark":"FLEURS","language":"Arabic","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, Voxtral Mini had a word error rate of 25.40 percent on FLEURS Arabic.","quote":"Model ar nl en fr de hi it pt es ... Voxtral Mini 25.40 6.27 3.77 4.87 4.40 9.26 2.51 3.76 3.52","where_in_source":"Table 4 (FLEURS), column ar","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-fleurs-ar-25-40","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v2, CHiME6","value":25.5,"low":null,"high":null,"as_written":"25.5","unit":"percent","system":"Whisper Large V2","kind":"Open model","benchmark":"CHiME6","language":"English","condition":null,"year":2022,"conversion":"Language is taken from the test set name: CHiME-6 is an English corpus. The condition is not stated in the fact.","claim":"On the CHiME6 test set, Whisper Large V2 had a word error rate of 25.5 percent.","quote":"CHiME6 65.8 25.5 61.2 VoxPopuli En 17.9 7.3 59.2","where_in_source":"Table 2, effective robustness across datasets","fact_url":"https://yapapp.cc/facts/whisper-chime6-wer-25-5","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision","source_publisher":"arXiv (eess.AS)","source_year":"2022","source_url":"https://arxiv.org/abs/2212.04356","source_doi":null,"checked":"2026-10-16"},{"row":"Whisper large-v2, Hebrew","value":27.1,"low":null,"high":null,"as_written":"27.1","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Hebrew","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the Fleurs benchmark, the Whisper large-v2 model had a word error rate of 27.1 percent for Hebrew.","quote":"Whisper large-v2 4.5 12.5 4.2 3.0 21.9 32.9 9.7 13.8 8.3 15.4 102.7 88.9 27.1 21.5","where_in_source":"Table 13 (Fleurs), column Hebrew","fact_url":"https://yapapp.cc/facts/whisper-fleurs-hebrew-27-1","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-17"},{"row":"Whisper large-v2, Lithuanian","value":28.1,"low":null,"high":null,"as_written":"28.1","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Lithuanian","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 28.1 percent for Lithuanian.","quote":"Whisper large-v2 75.6 101.5 28.1 23.1 38.5 16.5 100.7 110.5 38.3 8.7 76.6 115.7 9.5 47.1","where_in_source":"Table 13 (FLEURS), Whisper large-v2 row, column Lithuanian","fact_url":"https://yapapp.cc/facts/whisper-fleurs-lithuanian-28-1","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-19"},{"row":"Whisper large-v3, FLEURS Hindi","value":28.87,"low":null,"high":null,"as_written":"28.87","unit":"percent","system":"Whisper large-v3","kind":"Open model","benchmark":"FLEURS","language":"Hindi","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's FLEURS table, Whisper large-v3 had a word error rate of 28.87 percent in Hindi.","quote":"Model ar nl en fr de hi it pt es Whisper large-v3 15.44 5.87 4.00 5.55 5.46 28.87 2.71 3.90 2.81 GPT-4o mini Transcribe 14.02 5.54 3.19 4.51 3.76 12.36 2.02 3.54 2.58 Gemini 2.5 Flash 25.25 6.20 4.64 6.17 4.74 6.76","where_in_source":"Table 4 (FLEURS), column hi","fact_url":"https://yapapp.cc/facts/voxtral-2025-whisper-large-v3-hindi-28-87","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-17"},{"row":"Canary-Qwen 2.5B, noise -5 dB","value":30.6,"low":null,"high":null,"as_written":"30.6","unit":"percent","system":"NVIDIA Canary Qwen 2.5B","kind":"Open model","benchmark":"LibriSpeech test-clean with white noise","language":"English","condition":"Noisy or distant","year":2025,"conversion":null,"claim":"NVIDIA Canary Qwen 2.5B had a word error rate of 30.60 percent on LibriSpeech test-clean with added white noise at -5 dB SNR.","quote":"WER on [Librispeech Test Clean](https://www.openslr.org/12) at different SNR (signal to noise ratio) levels of additive white noise | **Version** | **Model** | **SNR 10** | **SNR 5** | **SNR 0** | **SNR -5** | ... | 2.5.0       | Canary-Qwen-2.5B |    2.41%   |   4.08%   |   9.83%   |    30.60%  |","where_in_source":"White-noise table, SNR -5 dB column","fact_url":"https://yapapp.cc/facts/nvidia-canary-qwen-2-5b-white-noise-snr-minus-5-db-30-60","source_title":"canary-qwen-2.5b model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/canary-qwen-2.5b/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v2, CS-Dialogue","value":31.11,"low":null,"high":null,"as_written":"31.11","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"CS-Dialogue","language":"Mandarin-English code-switching","condition":"Conversational","year":2025,"conversion":null,"claim":"On the CS-Dialogue Mandarin-English test set, Whisper Large-V2 had a 31.11 percent word error rate and a 15.29 percent mixed error rate, the highest error rates among the pre-trained models the authors tested.","quote":"exhibits the highest error rates among the pre-trained models, particularly in WER (31.11%) and MER (15.29%)","where_in_source":"Results, Table 8 and surrounding text","fact_url":"https://yapapp.cc/facts/zhou-2025-whisper-large-v2-cs-dialogue","source_title":"CS-Dialogue: A 104-Hour Dataset of Spontaneous Mandarin-English Code-Switching Dialogues for Speech Recognition","source_publisher":"arXiv","source_year":"2025","source_url":"https://arxiv.org/pdf/2502.18913","source_doi":null,"checked":"2026-10-17"},{"row":"Parakeet TDT v3, CoVoST 2 Slovenian","value":31.8,"low":null,"high":null,"as_written":"31.8","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"CoVoST 2","language":"Slovenian","condition":null,"year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 31.80 percent on the CoVoST 2 Slovenian test.","quote":"| Language | Fleurs | MLS | CoVoST | ... | **sl** | 24.03% | - | 31.80% |","where_in_source":"Multilingual ASR table, CoVoST column, sl row","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-covost-2-sl-31-80","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v2, Persian","value":32.9,"low":null,"high":null,"as_written":"32.9","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Persian","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the Fleurs benchmark, the Whisper large-v2 model had a word error rate of 32.9 percent for Persian.","quote":"Whisper large-v2 4.5 12.5 4.2 3.0 21.9 32.9 9.7 13.8 8.3 15.4 102.7 88.9 27.1 21.5","where_in_source":"Table 13 (Fleurs), column Persian","fact_url":"https://yapapp.cc/facts/whisper-fleurs-persian-32-9","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-17"},{"row":"Whisper large-v2, Welsh","value":33,"low":null,"high":null,"as_written":"33","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Welsh","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 33.0 percent for Welsh.","quote":"Whisper large-v2 36.7 140.3 16.0 106.2 23.4 45.4 14.6 104.1 15.7 7.3 14.7 13.3 33.0 13.8","where_in_source":"Table 13 (FLEURS), Whisper large-v2 row, column Welsh","fact_url":"https://yapapp.cc/facts/whisper-fleurs-welsh-33-0","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-19"},{"row":"Whisper large-v2, Serbian","value":33.9,"low":null,"high":null,"as_written":"33.9","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Serbian","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 33.9 percent for Serbian.","quote":"Whisper large-v2 6.7 75.3 102.4 5.4 93.7 4.3 14.4 5.6 156.5 11.7 23.1 121.0 102.9 33.9","where_in_source":"Table 13 (FLEURS), Whisper large-v2 row, column Serbian","fact_url":"https://yapapp.cc/facts/whisper-fleurs-serbian-33-9","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-19"},{"row":"Whisper large-v2, AMI SDM1","value":36.4,"low":null,"high":null,"as_written":"36.4","unit":"percent","system":"Whisper Large V2","kind":"Open model","benchmark":"AMI meetings (SDM1)","language":"English","condition":"Noisy or distant","year":2022,"conversion":"SDM1 is one single distant microphone. Language is taken from the test set name: AMI is an English meeting corpus.","claim":"On the AMI meeting corpus recorded with a single distant microphone (SDM1), Whisper Large V2 had a word error rate of 36.4 percent.","quote":"AMI SDM1 67.6 36.4 46.2 LibriSpeech Other 6.2 5.2 16.1","where_in_source":"Table 2, effective robustness across datasets, AMI SDM1 row","fact_url":"https://yapapp.cc/facts/whisper-ami-sdm1-36-4","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision","source_publisher":"arXiv (eess.AS)","source_year":"2022","source_url":"https://arxiv.org/abs/2212.04356","source_doi":null,"checked":"2026-10-17"},{"row":"Whisper large-v2, Afrikaans","value":36.7,"low":null,"high":null,"as_written":"36.7","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Afrikaans","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 36.7 percent for Afrikaans.","quote":"Whisper large-v2 36.7 140.3 16.0 106.2 23.4 45.4 14.6 104.1 15.7 7.3 14.7 13.3 33.0 13.8","where_in_source":"Table 13 (FLEURS), Whisper large-v2 row, column Afrikaans","fact_url":"https://yapapp.cc/facts/whisper-fleurs-afrikaans-36-7","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-19"},{"row":"Whisper large-v2, Kannada","value":37,"low":null,"high":null,"as_written":"37","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Kannada","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the Fleurs benchmark, the Whisper large-v2 model had a word error rate of 37.0 percent for Kannada.","quote":"Model Croatian Hungarian Armenian Indonesian Icelandic Italian Japanese Javanese Georgian Kazakh Khmer Kannada Korean Luxembourgish ... Whisper large-v2 13.4 17.0 44.6 7.1 38.2 4.0 5.3 nan 105.0 37.7 99.7 37.0 14.3 88.0","where_in_source":"Table 13 (Fleurs), column Kannada","fact_url":"https://yapapp.cc/facts/whisper-fleurs-kannada-37-0","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-17"},{"row":"Whisper large-v2, Kazakh","value":37.7,"low":null,"high":null,"as_written":"37.7","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Kazakh","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 37.7 percent for Kazakh.","quote":"Whisper large-v2 13.4 17.0 44.6 7.1 38.2 4.0 5.3 nan 105.0 37.7 99.7 37.0 14.3 88.0","where_in_source":"Table 13 (FLEURS), Whisper large-v2 row, column Kazakh","fact_url":"https://yapapp.cc/facts/whisper-fleurs-kazakh-37-7","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-19"},{"row":"Whisper large-v2, Icelandic","value":38.2,"low":null,"high":null,"as_written":"38.2","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Icelandic","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 38.2 percent for Icelandic.","quote":"Whisper large-v2 13.4 17.0 44.6 7.1 38.2 4.0 5.3 nan 105.0 37.7 99.7 37.0 14.3 88.0","where_in_source":"Table 13 (FLEURS), Whisper large-v2 row, column Icelandic","fact_url":"https://yapapp.cc/facts/whisper-fleurs-icelandic-38-2","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-19"},{"row":"Whisper large-v2, Marathi","value":38.3,"low":null,"high":null,"as_written":"38.3","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Marathi","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the Fleurs benchmark, the Whisper large-v2 model had a word error rate of 38.3 percent for Marathi.","quote":"Model Lingala Lao Lithuanian Latvian Maori Macedonian Malayalam Mongolian Marathi Malay ... Whisper large-v2 75.6 101.5 28.1 23.1 38.5 16.5 100.7 110.5 38.3 8.7","where_in_source":"Table 13 (Fleurs), column Marathi","fact_url":"https://yapapp.cc/facts/whisper-fleurs-marathi-38-3","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-17"},{"row":"Parakeet TDT v3, CoVoST 2 Latvian","value":38.36,"low":null,"high":null,"as_written":"38.36","unit":"percent","system":"NVIDIA Parakeet TDT 0.6B v3","kind":"Open model","benchmark":"CoVoST 2","language":"Latvian","condition":null,"year":2025,"conversion":null,"claim":"NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 38.36 percent on the CoVoST 2 Latvian test.","quote":"| Language | Fleurs | MLS | CoVoST | ... | **lv** | 22.84% | - | 38.36% |","where_in_source":"Multilingual ASR table, CoVoST column, lv row","fact_url":"https://yapapp.cc/facts/nvidia-parakeet-tdt-v3-covost-2-lv-38-36","source_title":"parakeet-tdt-0.6b-v3 model card (raw README)","source_publisher":"Hugging Face (NVIDIA model card)","source_year":"2025","source_url":"https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v2, Maori","value":38.5,"low":null,"high":null,"as_written":"38.5","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Maori","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 38.5 percent for Maori.","quote":"Whisper large-v2 75.6 101.5 28.1 23.1 38.5 16.5 100.7 110.5 38.3 8.7 76.6 115.7 9.5 47.1","where_in_source":"Table 13 (FLEURS), Whisper large-v2 row, column Maori","fact_url":"https://yapapp.cc/facts/whisper-fleurs-maori-38-5","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-19"},{"row":"Whisper large-v2, Swahili","value":39.3,"low":null,"high":null,"as_written":"39.3","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Swahili","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the Fleurs benchmark, the Whisper large-v2 model had a word error rate of 39.3 percent for Swahili.","quote":"Model Swedish Swahili Tamil Telugu Tajik Thai Turkish Ukrainian Urdu Uzbek Vietnamese Yoruba ... Whisper large-v2 8.5 39.3 17.5 99.0 85.8 11.5 8.4 8.6 22.6 90.2 10.3 94.8","where_in_source":"Table 13 (Fleurs)","fact_url":"https://yapapp.cc/facts/whisper-fleurs-swahili-39-3","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-16"},{"row":"Whisper large-v2, Common Voice Persian","value":39.4,"low":null,"high":null,"as_written":"39.4","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"Common Voice","language":"Persian","condition":null,"year":2022,"conversion":null,"claim":"On the Common Voice 9 test set, the Whisper large-v2 model had a word error rate of 39.4 percent for Persian.","quote":"Whisper large-v2 53.8 19.9 103.4 14.1 13.5 34.2 14.4 6.4 16.0 9.4 5.6 35.1 39.4","where_in_source":"Table 11 (CommonVoice9), column Persian","fact_url":"https://yapapp.cc/facts/whisper-cv9-persian-39-4","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-17"},{"row":"Whisper large-v2, Armenian","value":44.6,"low":null,"high":null,"as_written":"44.6","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Armenian","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 44.6 percent for Armenian.","quote":"Whisper large-v2 13.4 17.0 44.6 7.1 38.2 4.0 5.3 nan 105.0 37.7 99.7 37.0 14.3 88.0","where_in_source":"Table 13 (FLEURS), Whisper large-v2 row, column Armenian","fact_url":"https://yapapp.cc/facts/whisper-fleurs-armenian-44-6","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-19"},{"row":"Whisper large-v2, Belarusian","value":45.4,"low":null,"high":null,"as_written":"45.4","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Belarusian","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 45.4 percent for Belarusian.","quote":"Whisper large-v2 36.7 140.3 16.0 106.2 23.4 45.4 14.6 104.1 15.7 7.3 14.7 13.3 33.0 13.8","where_in_source":"Table 13 (FLEURS), Whisper large-v2 row, column Belarusian","fact_url":"https://yapapp.cc/facts/whisper-fleurs-belarusian-45-4","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-19"},{"row":"Whisper large-v3, Common Voice Hindi","value":46.75,"low":null,"high":null,"as_written":"46.75","unit":"percent","system":"Whisper large-v3","kind":"Open model","benchmark":"Common Voice","language":"Hindi","condition":null,"year":2025,"conversion":null,"claim":"On the Mozilla Common Voice Hindi test in the Voxtral paper, Whisper large-v3 had a word error rate of 46.75 percent.","quote":"Model ar nl en fr de hi it pt es Whisper large-v3 50.58 5.83 22.91 11.33 6.25 46.75 6.81 7.17 5.66","where_in_source":"Table 5 (Mozilla Common Voice), column hi","fact_url":"https://yapapp.cc/facts/voxtral-2025-whisper-large-v3-common-voice-hindi-46-75","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-17"},{"row":"Whisper large-v2, Nepali","value":47.1,"low":null,"high":null,"as_written":"47.1","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Nepali","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 47.1 percent for Nepali.","quote":"Whisper large-v2 75.6 101.5 28.1 23.1 38.5 16.5 100.7 110.5 38.3 8.7 76.6 115.7 9.5 47.1","where_in_source":"Table 13 (FLEURS), Whisper large-v2 row, column Nepali","fact_url":"https://yapapp.cc/facts/whisper-fleurs-nepali-47-1","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-19"},{"row":"Whisper large-v3, CV Arabic","value":50.58,"low":null,"high":null,"as_written":"50.58","unit":"percent","system":"Whisper large-v3","kind":"Open model","benchmark":"Common Voice","language":"Arabic","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, Whisper large-v3 had a word error rate of 50.58 percent on Common Voice Arabic.","quote":"Model ar nl en fr de hi it pt es ... Whisper large-v3 50.58 5.83 22.91 11.33 6.25 46.75 6.81 7.17 5.66","where_in_source":"Table 5 (Common Voice), column ar","fact_url":"https://yapapp.cc/facts/voxtral-2025-whisper-large-v3-common-voice-ar-50-58","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v2, Common Voice Arabic","value":53.8,"low":null,"high":null,"as_written":"53.8","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"Common Voice","language":"Arabic","condition":null,"year":2022,"conversion":null,"claim":"On the Common Voice 9 test set, the Whisper large-v2 model had a word error rate of 53.8 percent for Arabic.","quote":"Whisper large-v2 53.8 19.9 103.4 14.1 13.5 34.2 14.4 6.4 16.0 9.4 5.6 35.1 39.4","where_in_source":"Table 11 (CommonVoice9), column Arabic","fact_url":"https://yapapp.cc/facts/whisper-cv9-arabic-53-8","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-17"},{"row":"Voxtral Small, CV Arabic","value":61.97,"low":null,"high":null,"as_written":"61.97","unit":"percent","system":"Voxtral Small","kind":"Open model","benchmark":"Common Voice","language":"Arabic","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, Voxtral Small had a word error rate of 61.97 percent on Common Voice Arabic.","quote":"Model ar nl en fr de hi it pt es ... Voxtral Small 61.97 3.98 8.58 6.18 3.74 9.01 3.96 6.43 3.31","where_in_source":"Table 5 (Common Voice), column ar","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-small-common-voice-ar-61-97","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini Transcribe, CV Arabic","value":62.01,"low":null,"high":null,"as_written":"62.01","unit":"percent","system":"Voxtral Mini Transcribe","kind":"Open model","benchmark":"Common Voice","language":"Arabic","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, Voxtral Mini Transcribe had a word error rate of 62.01 percent on Common Voice Arabic.","quote":"Model ar nl en fr de hi it pt es ... Voxtral Mini Transcribe 62.01 4.71 8.25 7.29 4.85 10.42 4.37 6.70 3.96","where_in_source":"Table 5 (Common Voice), column ar","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-transcribe-common-voice-ar-62-01","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Voxtral Mini, CV Arabic","value":63.98,"low":null,"high":null,"as_written":"63.98","unit":"percent","system":"Voxtral Mini","kind":"Open model","benchmark":"Common Voice","language":"Arabic","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, Voxtral Mini had a word error rate of 63.98 percent on Common Voice Arabic.","quote":"Model ar nl en fr de hi it pt es ... Voxtral Mini 63.98 6.03 10.22 8.92 6.07 12.74 5.91 7.76 4.98","where_in_source":"Table 5 (Common Voice), column ar","fact_url":"https://yapapp.cc/facts/voxtral-2025-voxtral-mini-common-voice-ar-63-98","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper large-v2, Occitan","value":75.3,"low":null,"high":null,"as_written":"75.3","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Occitan","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 75.3 percent for Occitan.","quote":"Whisper large-v2 6.7 75.3 102.4 5.4 93.7 4.3 14.4 5.6 156.5 11.7 23.1 121.0 102.9 33.9","where_in_source":"Table 13 (FLEURS), Whisper large-v2 row, column Occitan","fact_url":"https://yapapp.cc/facts/whisper-fleurs-occitan-75-3","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-19"},{"row":"Whisper large-v2, Lingala","value":75.6,"low":null,"high":null,"as_written":"75.6","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Lingala","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 75.6 percent for Lingala.","quote":"Whisper large-v2 75.6 101.5 28.1 23.1 38.5 16.5 100.7 110.5 38.3 8.7 76.6 115.7 9.5 47.1","where_in_source":"Table 13 (FLEURS), Whisper large-v2 row, column Lingala","fact_url":"https://yapapp.cc/facts/whisper-fleurs-lingala-75-6","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-19"},{"row":"Whisper large-v2, Maltese","value":76.6,"low":null,"high":null,"as_written":"76.6","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Maltese","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 76.6 percent for Maltese.","quote":"Whisper large-v2 75.6 101.5 28.1 23.1 38.5 16.5 100.7 110.5 38.3 8.7 76.6 115.7 9.5 47.1","where_in_source":"Table 13 (FLEURS), Whisper large-v2 row, column Maltese","fact_url":"https://yapapp.cc/facts/whisper-fleurs-maltese-76-6","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-19"},{"row":"Whisper large-v2, Tajik","value":85.8,"low":null,"high":null,"as_written":"85.8","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Tajik","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 85.8 percent for Tajik.","quote":"Whisper large-v2 8.5 39.3 17.5 99.0 85.8 11.5 8.4 8.6 22.6 90.2 10.3 94.8","where_in_source":"Table 13 (FLEURS), Whisper large-v2 row, column Tajik","fact_url":"https://yapapp.cc/facts/whisper-fleurs-tajik-85-8","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-19"},{"row":"Whisper large-v2, Luxembourgish","value":88,"low":null,"high":null,"as_written":"88","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Luxembourgish","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 88.0 percent for Luxembourgish.","quote":"Whisper large-v2 13.4 17.0 44.6 7.1 38.2 4.0 5.3 nan 105.0 37.7 99.7 37.0 14.3 88.0","where_in_source":"Table 13 (FLEURS), Whisper large-v2 row, column Luxembourgish","fact_url":"https://yapapp.cc/facts/whisper-fleurs-luxembourgish-88-0","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-19"},{"row":"Whisper large-v2, Hausa","value":88.9,"low":null,"high":null,"as_written":"88.9","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Hausa","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS benchmark, Whisper large-v2 had a word error rate of 88.9 percent for Hausa.","quote":"Whisper large-v2 4.5 12.5 4.2 3.0 21.9 32.9 9.7 13.8 8.3 15.4 102.7 88.9 27.1 21.5","where_in_source":"Appendix D.2.4, Table 13","fact_url":"https://yapapp.cc/facts/whisper-fleurs-hausa-88-9","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-17"},{"row":"Whisper large-v2, Uzbek","value":90.2,"low":null,"high":null,"as_written":"90.2","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Uzbek","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 90.2 percent for Uzbek.","quote":"Whisper large-v2 8.5 39.3 17.5 99.0 85.8 11.5 8.4 8.6 22.6 90.2 10.3 94.8","where_in_source":"Table 13 (FLEURS), Whisper large-v2 row, column Uzbek","fact_url":"https://yapapp.cc/facts/whisper-fleurs-uzbek-90-2","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-19"},{"row":"Whisper large-v2, Pashto","value":93.7,"low":null,"high":null,"as_written":"93.7","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Pashto","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 93.7 percent for Pashto.","quote":"Whisper large-v2 6.7 75.3 102.4 5.4 93.7 4.3 14.4 5.6 156.5 11.7 23.1 121.0 102.9 33.9","where_in_source":"Table 13 (FLEURS), Whisper large-v2 row, column Pashto","fact_url":"https://yapapp.cc/facts/whisper-fleurs-pashto-93-7","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-19"},{"row":"Whisper large-v2, Yoruba","value":94.8,"low":null,"high":null,"as_written":"94.8","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Yoruba","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the Fleurs benchmark, the Whisper large-v2 model had a word error rate of 94.8 percent for Yoruba.","quote":"Model Swedish Swahili Tamil Telugu Tajik Thai Turkish Ukrainian Urdu Uzbek Vietnamese Yoruba ... Whisper large-v2 8.5 39.3 17.5 99.0 85.8 11.5 8.4 8.6 22.6 90.2 10.3 94.8","where_in_source":"Table 13 (Fleurs)","fact_url":"https://yapapp.cc/facts/whisper-fleurs-yoruba-94-8","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-16"},{"row":"Whisper large-v2, Telugu","value":99,"low":null,"high":null,"as_written":"99","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Telugu","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the Fleurs benchmark, the Whisper large-v2 model had a word error rate of 99.0 percent for Telugu.","quote":"Model Swedish Swahili Tamil Telugu Tajik Thai Turkish Ukrainian Urdu Uzbek Vietnamese Yoruba ... Whisper large-v2 8.5 39.3 17.5 99.0 85.8 11.5 8.4 8.6 22.6 90.2 10.3 94.8","where_in_source":"Table 13 (Fleurs)","fact_url":"https://yapapp.cc/facts/whisper-fleurs-telugu-99-0","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-16"},{"row":"Whisper large-v2, Khmer","value":99.7,"low":null,"high":null,"as_written":"99.7","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Khmer","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 99.7 percent for Khmer.","quote":"Whisper large-v2 13.4 17.0 44.6 7.1 38.2 4.0 5.3 nan 105.0 37.7 99.7 37.0 14.3 88.0","where_in_source":"Table 13 (FLEURS), Whisper large-v2 row, column Khmer","fact_url":"https://yapapp.cc/facts/whisper-fleurs-khmer-99-7","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-19"},{"row":"Whisper large-v2, Malayalam","value":100.7,"low":null,"high":null,"as_written":"100.7","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Malayalam","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the Fleurs benchmark, the Whisper large-v2 model had a word error rate of 100.7 percent for Malayalam.","quote":"Model Lingala Lao Lithuanian Latvian Maori Macedonian Malayalam Mongolian Marathi Malay ... Whisper large-v2 75.6 101.5 28.1 23.1 38.5 16.5 100.7 110.5 38.3 8.7","where_in_source":"Table 13 (Fleurs), column Malayalam","fact_url":"https://yapapp.cc/facts/whisper-fleurs-malayalam-100-7","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-17"},{"row":"Whisper large-v2, Punjabi","value":102.4,"low":null,"high":null,"as_written":"102.4","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Punjabi","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 102.4 percent for Punjabi.","quote":"Whisper large-v2 6.7 75.3 102.4 5.4 93.7 4.3 14.4 5.6 156.5 11.7 23.1 121.0 102.9 33.9","where_in_source":"Table 13 (FLEURS), Whisper large-v2 row, column Punjabi","fact_url":"https://yapapp.cc/facts/whisper-fleurs-punjabi-102-4","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-19"},{"row":"Whisper large-v2, Gujarati","value":102.7,"low":null,"high":null,"as_written":"102.7","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Gujarati","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 102.7 percent for Gujarati.","quote":"Whisper large-v2 4.5 12.5 4.2 3.0 21.9 32.9 9.7 13.8 8.3 15.4 102.7 88.9 27.1 21.5","where_in_source":"Table 13 (FLEURS), Whisper large-v2 row, column Gujarati","fact_url":"https://yapapp.cc/facts/whisper-fleurs-gujarati-102-7","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-19"},{"row":"Whisper large-v2, Somali","value":102.9,"low":null,"high":null,"as_written":"102.9","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Somali","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 102.9 percent for Somali.","quote":"Whisper large-v2 6.7 75.3 102.4 5.4 93.7 4.3 14.4 5.6 156.5 11.7 23.1 121.0 102.9 33.9","where_in_source":"Table 13 (FLEURS), Whisper large-v2 row, column Somali","fact_url":"https://yapapp.cc/facts/whisper-fleurs-somali-102-9","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-19"},{"row":"Whisper large-v2, Bengali","value":104.1,"low":null,"high":null,"as_written":"104.1","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Bengali","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the Fleurs benchmark, the Whisper large-v2 model had a word error rate of 104.1 percent for Bengali.","quote":"Model Afrikaans Amharic Arabic Assamese Azerbaijani Belarusian Bulgarian Bengali Bosnian Catalan Chinese Czech Welsh Danish ... Whisper large-v2 36.7 140.3 16.0 106.2 23.4 45.4 14.6 104.1 15.7","where_in_source":"Table 13 (Fleurs), column Bengali","fact_url":"https://yapapp.cc/facts/whisper-fleurs-bengali-104-1","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-17"},{"row":"Whisper large-v2, Georgian","value":105,"low":null,"high":null,"as_written":"105","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Georgian","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 105.0 percent for Georgian.","quote":"Whisper large-v2 13.4 17.0 44.6 7.1 38.2 4.0 5.3 nan 105.0 37.7 99.7 37.0 14.3 88.0","where_in_source":"Table 13 (FLEURS), Whisper large-v2 row, column Georgian","fact_url":"https://yapapp.cc/facts/whisper-fleurs-georgian-105-0","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-19"},{"row":"Whisper large-v2, Assamese","value":106.2,"low":null,"high":null,"as_written":"106.2","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Assamese","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 106.2 percent for Assamese.","quote":"Whisper large-v2 36.7 140.3 16.0 106.2 23.4 45.4 14.6 104.1 15.7 7.3 14.7 13.3 33.0 13.8","where_in_source":"Table 13 (FLEURS), Whisper large-v2 row, column Assamese","fact_url":"https://yapapp.cc/facts/whisper-fleurs-assamese-106-2","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-19"},{"row":"Whisper large-v2, Mongolian","value":110.5,"low":null,"high":null,"as_written":"110.5","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Mongolian","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 110.5 percent for Mongolian.","quote":"Whisper large-v2 75.6 101.5 28.1 23.1 38.5 16.5 100.7 110.5 38.3 8.7 76.6 115.7 9.5 47.1","where_in_source":"Table 13 (FLEURS), Whisper large-v2 row, column Mongolian","fact_url":"https://yapapp.cc/facts/whisper-fleurs-mongolian-110-5","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-19"},{"row":"Whisper large-v2, Shona","value":121,"low":null,"high":null,"as_written":"121","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Shona","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 121.0 percent for Shona.","quote":"Whisper large-v2 6.7 75.3 102.4 5.4 93.7 4.3 14.4 5.6 156.5 11.7 23.1 121.0 102.9 33.9","where_in_source":"Table 13 (FLEURS), Whisper large-v2 row, column Shona","fact_url":"https://yapapp.cc/facts/whisper-fleurs-shona-121-0","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-19"},{"row":"Whisper large-v2, Amharic","value":140.3,"low":null,"high":null,"as_written":"140.3","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Amharic","condition":null,"year":2022,"conversion":null,"claim":"On the FLEURS benchmark, Whisper large-v2 had a word error rate of 140.3 percent for Amharic, meaning more errors than words in the reference.","quote":"Whisper large-v2 36.7 140.3 16.0 106.2 23.4 45.4 14.6 104.1 15.7 7.3 14.7 13.3 33.0 13.8","where_in_source":"Appendix D.2.4, Table 13","fact_url":"https://yapapp.cc/facts/whisper-fleurs-amharic-140-3","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-17"},{"row":"Whisper large-v2, Sindhi","value":156.5,"low":null,"high":null,"as_written":"156.5","unit":"percent","system":"Whisper large-v2","kind":"Open model","benchmark":"FLEURS","language":"Sindhi","condition":"Clean read speech","year":2022,"conversion":null,"claim":"On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 156.5 percent for Sindhi.","quote":"Whisper large-v2 6.7 75.3 102.4 5.4 93.7 4.3 14.4 5.6 156.5 11.7 23.1 121.0 102.9 33.9","where_in_source":"Table 13 (FLEURS), Whisper large-v2 row, column Sindhi","fact_url":"https://yapapp.cc/facts/whisper-fleurs-sindhi-156-5","source_title":"Robust Speech Recognition via Large-Scale Weak Supervision (full text)","source_publisher":"OpenAI (arXiv)","source_year":"2022","source_url":"https://arxiv.org/pdf/2212.04356","source_doi":"10.48550/arXiv.2212.04356","checked":"2026-10-19"},{"row":"ElevenLabs Scribe, FLEURS Italian","value":1.48,"low":null,"high":null,"as_written":"1.48","unit":"percent","system":"ElevenLabs Scribe","kind":"Cloud service","benchmark":"FLEURS","language":"Italian","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, ElevenLabs Scribe had a word error rate of 1.48 percent on FLEURS Italian.","quote":"Model ar nl en fr de hi it pt es ... Scribe 11.58 4.63 3.29 5.07 4.78 5.67 1.48 4.50 3.13","where_in_source":"Table 4 (FLEURS), column it","fact_url":"https://yapapp.cc/facts/voxtral-2025-elevenlabs-scribe-fleurs-it-1-48","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Fun-Realtime-ASR-preview Alibaba Cloud","value":1.7,"low":null,"high":null,"as_written":"1.7","unit":"percent","system":"Fun-Realtime-ASR-preview Alibaba Cloud","kind":"Cloud service","benchmark":"AA-WER","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Language is not named on the leaderboard page; the AA-WER v2.0 article describes a normalizer built on Whisper's English normalizer.","claim":"On Artificial Analysis's speech-to-text leaderboard, Fun-Realtime-ASR-preview Alibaba Cloud had an AA-WER of 1.7 percent, its word error rate on the index's three test datasets.","quote":"Fun-Realtime-ASR-preview Alibaba Cloud 1.7% 20.4 0.00 Details","where_in_source":"Summary of Key Metrics table","fact_url":"https://yapapp.cc/facts/aa-2026-fun-realtime-asr-preview-alibaba-cloud-aa-wer","source_title":"Speech to Text (ASR) Providers Leaderboard & Comparison","source_publisher":"Artificial Analysis","source_year":"2026","source_url":"https://artificialanalysis.ai/speech-to-text","source_doi":null,"checked":"2026-10-19"},{"row":"ElevenLabs Scribe, LS test-clean","value":1.8,"low":null,"high":null,"as_written":"1.8","unit":"percent","system":"ElevenLabs Scribe","kind":"Cloud service","benchmark":"LibriSpeech test-clean","language":"English","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, ElevenLabs Scribe had a word error rate of 1.80 percent on LibriSpeech test-clean.","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... ElevenLabs Scribe 1.80 3.44 10.52 6.95 10.62 8.35 3.16 7.39 9.16","where_in_source":"Table 3 (short-form English), LibriSpeech test-clean column","fact_url":"https://yapapp.cc/facts/voxtral-2025-elevenlabs-scribe-librispeech-test-clean-1-80","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"GPT-4o mini Transcribe, LS test-clean","value":1.92,"low":null,"high":null,"as_written":"1.92","unit":"percent","system":"GPT-4o mini Transcribe","kind":"Cloud service","benchmark":"LibriSpeech test-clean","language":"English","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, GPT-4o mini Transcribe had a word error rate of 1.92 percent on LibriSpeech test-clean.","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... GPT4o mini Transcribe 1.92 4.70 14.80 7.34 17.31 11.35 4.51 10.09 12.27","where_in_source":"Table 3 (short-form English), LibriSpeech test-clean column","fact_url":"https://yapapp.cc/facts/voxtral-2025-gpt-4o-mini-transcribe-librispeech-test-clean-1-92","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"MAI-Transcribe-2 Microsoft AI","value":2,"low":null,"high":null,"as_written":"2","unit":"percent","system":"MAI-Transcribe-2 Microsoft AI","kind":"Cloud service","benchmark":"AA-WER","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Language is not named on the leaderboard page; the AA-WER v2.0 article describes a normalizer built on Whisper's English normalizer.","claim":"On Artificial Analysis's speech-to-text leaderboard, MAI-Transcribe-2 Microsoft AI had an AA-WER of 2.0 percent, its word error rate on the index's three test datasets.","quote":"MAI-Transcribe-2 Microsoft AI 2.0% 317.0 1.67 Details","where_in_source":"Summary of Key Metrics table","fact_url":"https://yapapp.cc/facts/aa-2026-mai-transcribe-2-microsoft-ai-aa-wer","source_title":"Speech to Text (ASR) Providers Leaderboard & Comparison","source_publisher":"Artificial Analysis","source_year":"2026","source_url":"https://artificialanalysis.ai/speech-to-text","source_doi":null,"checked":"2026-10-19"},{"row":"GPT-4o mini Transcribe, FLEURS Italian","value":2.02,"low":null,"high":null,"as_written":"2.02","unit":"percent","system":"GPT-4o mini Transcribe","kind":"Cloud service","benchmark":"FLEURS","language":"Italian","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, GPT-4o mini Transcribe had a word error rate of 2.02 percent on FLEURS Italian.","quote":"Model ar nl en fr de hi it pt es ... GPT-4o mini Transcribe 14.02 5.54 3.19 4.51 3.76 12.36 2.02 3.54 2.58","where_in_source":"Table 4 (FLEURS), column it","fact_url":"https://yapapp.cc/facts/voxtral-2025-gpt-4o-mini-transcribe-fleurs-it-2-02","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Scribe v2 ElevenLabs","value":2.2,"low":null,"high":null,"as_written":"2.2","unit":"percent","system":"Scribe v2 ElevenLabs","kind":"Cloud service","benchmark":"AA-WER","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Language is not named on the leaderboard page; the AA-WER v2.0 article describes a normalizer built on Whisper's English normalizer.","claim":"On Artificial Analysis's speech-to-text leaderboard, Scribe v2 ElevenLabs had an AA-WER of 2.2 percent, its word error rate on the index's three test datasets.","quote":"Scribe v2 ElevenLabs 2.2% 63.0 3.67 Details","where_in_source":"Summary of Key Metrics table","fact_url":"https://yapapp.cc/facts/aa-2026-scribe-v2-elevenlabs-aa-wer","source_title":"Speech to Text (ASR) Providers Leaderboard & Comparison","source_publisher":"Artificial Analysis","source_year":"2026","source_url":"https://artificialanalysis.ai/speech-to-text","source_doi":null,"checked":"2026-10-19"},{"row":"Gemini 2.5 Flash, FLEURS Italian","value":2.21,"low":null,"high":null,"as_written":"2.21","unit":"percent","system":"Gemini 2.5 Flash","kind":"Cloud service","benchmark":"FLEURS","language":"Italian","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, Gemini 2.5 Flash had a word error rate of 2.21 percent on FLEURS Italian.","quote":"Model ar nl en fr de hi it pt es ... Gemini 2.5 Flash 25.25 6.20 4.64 6.17 4.74 6.76 2.21 4.23 3.17","where_in_source":"Table 4 (FLEURS), column it","fact_url":"https://yapapp.cc/facts/voxtral-2025-gemini-2-5-flash-fleurs-it-2-21","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Grok Voice Transcribe 2.0 SpaceXAI","value":2.3,"low":null,"high":null,"as_written":"2.3","unit":"percent","system":"Grok Voice Transcribe 2.0 SpaceXAI","kind":"Cloud service","benchmark":"AA-WER","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Language is not named on the leaderboard page; the AA-WER v2.0 article describes a normalizer built on Whisper's English normalizer.","claim":"On Artificial Analysis's speech-to-text leaderboard, Grok Voice Transcribe 2.0 SpaceXAI had an AA-WER of 2.3 percent, its word error rate on the index's three test datasets.","quote":"Grok Voice Transcribe 2.0 SpaceXAI 2.3% 158.5 1.67 Details","where_in_source":"Summary of Key Metrics table","fact_url":"https://yapapp.cc/facts/aa-2026-grok-voice-transcribe-2-0-spacexai-aa-wer","source_title":"Speech to Text (ASR) Providers Leaderboard & Comparison","source_publisher":"Artificial Analysis","source_year":"2026","source_url":"https://artificialanalysis.ai/speech-to-text","source_doi":null,"checked":"2026-10-19"},{"row":"ElevenLabs Scribe, CV Dutch","value":2.38,"low":null,"high":null,"as_written":"2.38","unit":"percent","system":"ElevenLabs Scribe","kind":"Cloud service","benchmark":"Common Voice","language":"Dutch","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, ElevenLabs Scribe had a word error rate of 2.38 percent on Common Voice Dutch.","quote":"Model ar nl en fr de hi it pt es ... Scribe 47.03 2.38 6.59 5.44 3.52 17.29 2.99 5.46 3.27","where_in_source":"Table 5 (Common Voice), column nl","fact_url":"https://yapapp.cc/facts/voxtral-2025-elevenlabs-scribe-common-voice-nl-2-38","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"MAI-Transcribe-1.5 Microsoft AI","value":2.4,"low":null,"high":null,"as_written":"2.4","unit":"percent","system":"MAI-Transcribe-1.5 Microsoft AI","kind":"Cloud service","benchmark":"AA-WER","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Language is not named on the leaderboard page; the AA-WER v2.0 article describes a normalizer built on Whisper's English normalizer.","claim":"On Artificial Analysis's speech-to-text leaderboard, MAI-Transcribe-1.5 Microsoft AI had an AA-WER of 2.4 percent, its word error rate on the index's three test datasets.","quote":"MAI-Transcribe-1.5 Microsoft AI 2.4% 80.5 6.00 Details","where_in_source":"Summary of Key Metrics table","fact_url":"https://yapapp.cc/facts/aa-2026-mai-transcribe-1-5-microsoft-ai-aa-wer","source_title":"Speech to Text (ASR) Providers Leaderboard & Comparison","source_publisher":"Artificial Analysis","source_year":"2026","source_url":"https://artificialanalysis.ai/speech-to-text","source_doi":null,"checked":"2026-10-19"},{"row":"StepAudio 3 ASR, StepFun StepFun","value":2.5,"low":null,"high":null,"as_written":"2.5","unit":"percent","system":"StepAudio 3 ASR, StepFun StepFun","kind":"Cloud service","benchmark":"AA-WER","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Language is not named on the leaderboard page; the AA-WER v2.0 article describes a normalizer built on Whisper's English normalizer.","claim":"On Artificial Analysis's speech-to-text leaderboard, StepAudio 3 ASR, StepFun StepFun had an AA-WER of 2.5 percent, its word error rate on the index's three test datasets.","quote":"StepAudio 3 ASR, StepFun StepFun 2.5% 79.4 6.67 Details","where_in_source":"Summary of Key Metrics table","fact_url":"https://yapapp.cc/facts/aa-2026-stepaudio-3-asr-stepfun-stepfun-aa-wer","source_title":"Speech to Text (ASR) Providers Leaderboard & Comparison","source_publisher":"Artificial Analysis","source_year":"2026","source_url":"https://artificialanalysis.ai/speech-to-text","source_doi":null,"checked":"2026-10-19"},{"row":"GPT-4o mini Transcribe, FLEURS Spanish","value":2.58,"low":null,"high":null,"as_written":"2.58","unit":"percent","system":"GPT-4o mini Transcribe","kind":"Cloud service","benchmark":"FLEURS","language":"Spanish","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, GPT-4o mini Transcribe had a word error rate of 2.58 percent on FLEURS Spanish.","quote":"Model ar nl en fr de hi it pt es ... GPT-4o mini Transcribe 14.02 5.54 3.19 4.51 3.76 12.36 2.02 3.54 2.58","where_in_source":"Table 4 (FLEURS), column es","fact_url":"https://yapapp.cc/facts/voxtral-2025-gpt-4o-mini-transcribe-fleurs-es-2-58","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Gemini 3.5 Transcribe Google","value":2.6,"low":null,"high":null,"as_written":"2.6","unit":"percent","system":"Gemini 3.5 Transcribe Google","kind":"Cloud service","benchmark":"AA-WER","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Language is not named on the leaderboard page; the AA-WER v2.0 article describes a normalizer built on Whisper's English normalizer.","claim":"On Artificial Analysis's speech-to-text leaderboard, Gemini 3.5 Transcribe Google had an AA-WER of 2.6 percent, its word error rate on the index's three test datasets.","quote":"Gemini 3.5 Transcribe Google 2.6% 91.7 5.00 Details","where_in_source":"Summary of Key Metrics table","fact_url":"https://yapapp.cc/facts/aa-2026-gemini-3-5-transcribe-google-aa-wer","source_title":"Speech to Text (ASR) Providers Leaderboard & Comparison","source_publisher":"Artificial Analysis","source_year":"2026","source_url":"https://artificialanalysis.ai/speech-to-text","source_doi":null,"checked":"2026-10-19"},{"row":"Gemini 3.1 Pro Preview (High) Google","value":2.8,"low":null,"high":null,"as_written":"2.8","unit":"percent","system":"Gemini 3.1 Pro Preview (High) Google","kind":"Cloud service","benchmark":"AA-WER","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Language is not named on the leaderboard page; the AA-WER v2.0 article describes a normalizer built on Whisper's English normalizer.","claim":"On Artificial Analysis's speech-to-text leaderboard, Gemini 3.1 Pro Preview (High) Google had an AA-WER of 2.8 percent, its word error rate on the index's three test datasets.","quote":"Gemini 3.1 Pro Preview (High) Google 2.8% 7.8 18.15 Details","where_in_source":"Summary of Key Metrics table","fact_url":"https://yapapp.cc/facts/aa-2026-gemini-3-1-pro-preview-high-google-aa-wer","source_title":"Speech to Text (ASR) Providers Leaderboard & Comparison","source_publisher":"Artificial Analysis","source_year":"2026","source_url":"https://artificialanalysis.ai/speech-to-text","source_doi":null,"checked":"2026-10-19"},{"row":"Gemini 2.5 Pro Google","value":2.9,"low":null,"high":null,"as_written":"2.9","unit":"percent","system":"Gemini 2.5 Pro Google","kind":"Cloud service","benchmark":"AA-WER","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Language is not named on the leaderboard page; the AA-WER v2.0 article describes a normalizer built on Whisper's English normalizer.","claim":"On Artificial Analysis's speech-to-text leaderboard, Gemini 2.5 Pro Google had an AA-WER of 2.9 percent, its word error rate on the index's three test datasets.","quote":"Gemini 2.5 Pro Google 2.9% 10.6 11.39 Details","where_in_source":"Summary of Key Metrics table","fact_url":"https://yapapp.cc/facts/aa-2026-gemini-2-5-pro-google-aa-wer","source_title":"Speech to Text (ASR) Providers Leaderboard & Comparison","source_publisher":"Artificial Analysis","source_year":"2026","source_url":"https://artificialanalysis.ai/speech-to-text","source_doi":null,"checked":"2026-10-19"},{"row":"Gemini 2.5 Flash, LS test-clean English","value":2.97,"low":null,"high":null,"as_written":"2.97","unit":"percent","system":"Gemini 2.5 Flash","kind":"Cloud service","benchmark":"LibriSpeech test-clean","language":"English","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, Gemini 2.5 Flash had a word error rate of 2.97 percent on LibriSpeech test-clean.","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... Gemini 2.5 Flash 2.97 6.15 10.99 7.84 9.57 14.79 4.00 8.09 10.80","where_in_source":"Table 3 (short-form English), LibriSpeech test-clean column","fact_url":"https://yapapp.cc/facts/voxtral-2025-gemini-2-5-flash-librispeech-test-clean-2-97","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"ElevenLabs Scribe, CV Italian","value":2.99,"low":null,"high":null,"as_written":"2.99","unit":"percent","system":"ElevenLabs Scribe","kind":"Cloud service","benchmark":"Common Voice","language":"Italian","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, ElevenLabs Scribe had a word error rate of 2.99 percent on Common Voice Italian.","quote":"Model ar nl en fr de hi it pt es ... Scribe 47.03 2.38 6.59 5.44 3.52 17.29 2.99 5.46 3.27","where_in_source":"Table 5 (Common Voice), column it","fact_url":"https://yapapp.cc/facts/voxtral-2025-elevenlabs-scribe-common-voice-it-2-99","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Universal-3.5 Pro AssemblyAI","value":3,"low":null,"high":null,"as_written":"3","unit":"percent","system":"Universal-3.5 Pro AssemblyAI","kind":"Cloud service","benchmark":"AA-WER","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Language is not named on the leaderboard page; the AA-WER v2.0 article describes a normalizer built on Whisper's English normalizer.","claim":"On Artificial Analysis's speech-to-text leaderboard, Universal-3.5 Pro AssemblyAI had an AA-WER of 3.0 percent, its word error rate on the index's three test datasets.","quote":"Universal-3.5 Pro AssemblyAI 3.0% 105.5 3.50 Details","where_in_source":"Summary of Key Metrics table","fact_url":"https://yapapp.cc/facts/aa-2026-universal-3-5-pro-assemblyai-aa-wer","source_title":"Speech to Text (ASR) Providers Leaderboard & Comparison","source_publisher":"Artificial Analysis","source_year":"2026","source_url":"https://artificialanalysis.ai/speech-to-text","source_doi":null,"checked":"2026-10-19"},{"row":"Universal-3 Pro AssemblyAI","value":3.1,"low":null,"high":null,"as_written":"3.1","unit":"percent","system":"Universal-3 Pro AssemblyAI","kind":"Cloud service","benchmark":"AA-WER","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Language is not named on the leaderboard page; the AA-WER v2.0 article describes a normalizer built on Whisper's English normalizer.","claim":"On Artificial Analysis's speech-to-text leaderboard, Universal-3 Pro AssemblyAI had an AA-WER of 3.1 percent, its word error rate on the index's three test datasets.","quote":"Universal-3 Pro AssemblyAI 3.1% 109.9 3.50 Details","where_in_source":"Summary of Key Metrics table","fact_url":"https://yapapp.cc/facts/aa-2026-universal-3-pro-assemblyai-aa-wer","source_title":"Speech to Text (ASR) Providers Leaderboard & Comparison","source_publisher":"Artificial Analysis","source_year":"2026","source_url":"https://artificialanalysis.ai/speech-to-text","source_doi":null,"checked":"2026-10-19"},{"row":"ElevenLabs Scribe, FLEURS Spanish","value":3.13,"low":null,"high":null,"as_written":"3.13","unit":"percent","system":"ElevenLabs Scribe","kind":"Cloud service","benchmark":"FLEURS","language":"Spanish","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, ElevenLabs Scribe had a word error rate of 3.13 percent on FLEURS Spanish.","quote":"Model ar nl en fr de hi it pt es ... Scribe 11.58 4.63 3.29 5.07 4.78 5.67 1.48 4.50 3.13","where_in_source":"Table 4 (FLEURS), column es","fact_url":"https://yapapp.cc/facts/voxtral-2025-elevenlabs-scribe-fleurs-es-3-13","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"ElevenLabs Scribe, SPGI English","value":3.16,"low":null,"high":null,"as_written":"3.16","unit":"percent","system":"ElevenLabs Scribe","kind":"Cloud service","benchmark":"SPGI Speech","language":"English","condition":"Specialist vocabulary","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, ElevenLabs Scribe had a word error rate of 3.16 percent on SPGI Speech.","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... ElevenLabs Scribe 1.80 3.44 10.52 6.95 10.62 8.35 3.16 7.39 9.16","where_in_source":"Table 3 (short-form English), SPGI Speech column","fact_url":"https://yapapp.cc/facts/voxtral-2025-elevenlabs-scribe-spgi-speech-3-16","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Gemini 2.5 Flash, FLEURS Spanish","value":3.17,"low":null,"high":null,"as_written":"3.17","unit":"percent","system":"Gemini 2.5 Flash","kind":"Cloud service","benchmark":"FLEURS","language":"Spanish","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, Gemini 2.5 Flash had a word error rate of 3.17 percent on FLEURS Spanish.","quote":"Model ar nl en fr de hi it pt es ... Gemini 2.5 Flash 25.25 6.20 4.64 6.17 4.74 6.76 2.21 4.23 3.17","where_in_source":"Table 4 (FLEURS), column es","fact_url":"https://yapapp.cc/facts/voxtral-2025-gemini-2-5-flash-fleurs-es-3-17","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"GPT-4o mini Transcribe, FLEURS English","value":3.19,"low":null,"high":null,"as_written":"3.19","unit":"percent","system":"GPT-4o mini Transcribe","kind":"Cloud service","benchmark":"FLEURS","language":"English","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, GPT-4o mini Transcribe had a word error rate of 3.19 percent on FLEURS English.","quote":"Model ar nl en fr de hi it pt es ... GPT-4o mini Transcribe 14.02 5.54 3.19 4.51 3.76 12.36 2.02 3.54 2.58","where_in_source":"Table 4 (FLEURS), column en","fact_url":"https://yapapp.cc/facts/voxtral-2025-gpt-4o-mini-transcribe-fleurs-en-3-19","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Solaria-3, Gladia Gladia","value":3.2,"low":null,"high":null,"as_written":"3.2","unit":"percent","system":"Solaria-3, Gladia Gladia","kind":"Cloud service","benchmark":"AA-WER","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Language is not named on the leaderboard page; the AA-WER v2.0 article describes a normalizer built on Whisper's English normalizer.","claim":"On Artificial Analysis's speech-to-text leaderboard, Solaria-3, Gladia Gladia had an AA-WER of 3.2 percent, its word error rate on the index's three test datasets.","quote":"Solaria-3, Gladia Gladia 3.2% 91.8 10.16 Details","where_in_source":"Summary of Key Metrics table","fact_url":"https://yapapp.cc/facts/aa-2026-solaria-3-gladia-gladia-aa-wer","source_title":"Speech to Text (ASR) Providers Leaderboard & Comparison","source_publisher":"Artificial Analysis","source_year":"2026","source_url":"https://artificialanalysis.ai/speech-to-text","source_doi":null,"checked":"2026-10-19"},{"row":"ElevenLabs Scribe, CV Spanish","value":3.27,"low":null,"high":null,"as_written":"3.27","unit":"percent","system":"ElevenLabs Scribe","kind":"Cloud service","benchmark":"Common Voice","language":"Spanish","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, ElevenLabs Scribe had a word error rate of 3.27 percent on Common Voice Spanish.","quote":"Model ar nl en fr de hi it pt es ... Scribe 47.03 2.38 6.59 5.44 3.52 17.29 2.99 5.46 3.27","where_in_source":"Table 5 (Common Voice), column es","fact_url":"https://yapapp.cc/facts/voxtral-2025-elevenlabs-scribe-common-voice-es-3-27","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"ElevenLabs Scribe, FLEURS English","value":3.29,"low":null,"high":null,"as_written":"3.29","unit":"percent","system":"ElevenLabs Scribe","kind":"Cloud service","benchmark":"FLEURS","language":"English","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, ElevenLabs Scribe had a word error rate of 3.29 percent on FLEURS English.","quote":"Model ar nl en fr de hi it pt es ... Scribe 11.58 4.63 3.29 5.07 4.78 5.67 1.48 4.50 3.13","where_in_source":"Table 4 (FLEURS), column en","fact_url":"https://yapapp.cc/facts/voxtral-2025-elevenlabs-scribe-fleurs-en-3-29","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"GPT Transcribe, OpenAI OpenAI","value":3.3,"low":null,"high":null,"as_written":"3.3","unit":"percent","system":"GPT Transcribe, OpenAI OpenAI","kind":"Cloud service","benchmark":"AA-WER","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Language is not named on the leaderboard page; the AA-WER v2.0 article describes a normalizer built on Whisper's English normalizer.","claim":"On Artificial Analysis's speech-to-text leaderboard, GPT Transcribe, OpenAI OpenAI had an AA-WER of 3.3 percent, its word error rate on the index's three test datasets.","quote":"GPT Transcribe, OpenAI OpenAI 3.3% 56.3 4.50 Details","where_in_source":"Summary of Key Metrics table","fact_url":"https://yapapp.cc/facts/aa-2026-gpt-transcribe-openai-openai-aa-wer","source_title":"Speech to Text (ASR) Providers Leaderboard & Comparison","source_publisher":"Artificial Analysis","source_year":"2026","source_url":"https://artificialanalysis.ai/speech-to-text","source_doi":null,"checked":"2026-10-19"},{"row":"Gemini 3.1 Flash-Lite Preview (Minimal)","value":3.4,"low":null,"high":null,"as_written":"3.4","unit":"percent","system":"Gemini 3.1 Flash-Lite Preview (Minimal) Google","kind":"Cloud service","benchmark":"AA-WER","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Language is not named on the leaderboard page; the AA-WER v2.0 article describes a normalizer built on Whisper's English normalizer.","claim":"On Artificial Analysis's speech-to-text leaderboard, Gemini 3.1 Flash-Lite Preview (Minimal) Google had an AA-WER of 3.4 percent, its word error rate on the index's three test datasets.","quote":"Gemini 3.1 Flash-Lite Preview (Minimal) Google 3.4% 88.5 5.83 Details","where_in_source":"Summary of Key Metrics table","fact_url":"https://yapapp.cc/facts/aa-2026-gemini-3-1-flash-lite-preview-minimal-google-aa-wer","source_title":"Speech to Text (ASR) Providers Leaderboard & Comparison","source_publisher":"Artificial Analysis","source_year":"2026","source_url":"https://artificialanalysis.ai/speech-to-text","source_doi":null,"checked":"2026-10-19"},{"row":"ElevenLabs Scribe, LS test-other","value":3.44,"low":null,"high":null,"as_written":"3.44","unit":"percent","system":"ElevenLabs Scribe","kind":"Cloud service","benchmark":"LibriSpeech test-other","language":"English","condition":"Mixed benchmark","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, ElevenLabs Scribe had a word error rate of 3.44 percent on LibriSpeech test-other.","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... ElevenLabs Scribe 1.80 3.44 10.52 6.95 10.62 8.35 3.16 7.39 9.16","where_in_source":"Table 3 (short-form English), LibriSpeech test-other column","fact_url":"https://yapapp.cc/facts/voxtral-2025-elevenlabs-scribe-librispeech-test-other-3-44","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Qwen3.5 Omni Plus Alibaba Cloud","value":3.5,"low":null,"high":null,"as_written":"3.5","unit":"percent","system":"Qwen3.5 Omni Plus Alibaba Cloud","kind":"Cloud service","benchmark":"AA-WER","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Language is not named on the leaderboard page; the AA-WER v2.0 article describes a normalizer built on Whisper's English normalizer.","claim":"On Artificial Analysis's speech-to-text leaderboard, Qwen3.5 Omni Plus Alibaba Cloud had an AA-WER of 3.5 percent, its word error rate on the index's three test datasets.","quote":"Qwen3.5 Omni Plus Alibaba Cloud 3.5% 166.7 0.00 Details","where_in_source":"Summary of Key Metrics table","fact_url":"https://yapapp.cc/facts/aa-2026-qwen3-5-omni-plus-alibaba-cloud-aa-wer","source_title":"Speech to Text (ASR) Providers Leaderboard & Comparison","source_publisher":"Artificial Analysis","source_year":"2026","source_url":"https://artificialanalysis.ai/speech-to-text","source_doi":null,"checked":"2026-10-19"},{"row":"ElevenLabs Scribe, CV German","value":3.52,"low":null,"high":null,"as_written":"3.52","unit":"percent","system":"ElevenLabs Scribe","kind":"Cloud service","benchmark":"Common Voice","language":"German","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, ElevenLabs Scribe had a word error rate of 3.52 percent on Common Voice German.","quote":"Model ar nl en fr de hi it pt es ... Scribe 47.03 2.38 6.59 5.44 3.52 17.29 2.99 5.46 3.27","where_in_source":"Table 5 (Common Voice), column de","fact_url":"https://yapapp.cc/facts/voxtral-2025-elevenlabs-scribe-common-voice-de-3-52","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"GPT-4o mini Transcribe, FLEURS","value":3.54,"low":null,"high":null,"as_written":"3.54","unit":"percent","system":"GPT-4o mini Transcribe","kind":"Cloud service","benchmark":"FLEURS","language":"Portuguese","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, GPT-4o mini Transcribe had a word error rate of 3.54 percent on FLEURS Portuguese.","quote":"Model ar nl en fr de hi it pt es ... GPT-4o mini Transcribe 14.02 5.54 3.19 4.51 3.76 12.36 2.02 3.54 2.58","where_in_source":"Table 4 (FLEURS), column pt","fact_url":"https://yapapp.cc/facts/voxtral-2025-gpt-4o-mini-transcribe-fleurs-pt-3-54","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Gemini 3.1 Pro Preview (Low) Google","value":3.6,"low":null,"high":null,"as_written":"3.6","unit":"percent","system":"Gemini 3.1 Pro Preview (Low) Google","kind":"Cloud service","benchmark":"AA-WER","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Language is not named on the leaderboard page; the AA-WER v2.0 article describes a normalizer built on Whisper's English normalizer.","claim":"On Artificial Analysis's speech-to-text leaderboard, Gemini 3.1 Pro Preview (Low) Google had an AA-WER of 3.6 percent, its word error rate on the index's three test datasets.","quote":"Gemini 3.1 Pro Preview (Low) Google 3.6% 7.7 7.72 Details","where_in_source":"Summary of Key Metrics table","fact_url":"https://yapapp.cc/facts/aa-2026-gemini-3-1-pro-preview-low-google-aa-wer","source_title":"Speech to Text (ASR) Providers Leaderboard & Comparison","source_publisher":"Artificial Analysis","source_year":"2026","source_url":"https://artificialanalysis.ai/speech-to-text","source_doi":null,"checked":"2026-10-19"},{"row":"GPT-4o mini Transcribe, FLEURS German","value":3.76,"low":null,"high":null,"as_written":"3.76","unit":"percent","system":"GPT-4o mini Transcribe","kind":"Cloud service","benchmark":"FLEURS","language":"German","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, GPT-4o mini Transcribe had a word error rate of 3.76 percent on FLEURS German.","quote":"Model ar nl en fr de hi it pt es ... GPT-4o mini Transcribe 14.02 5.54 3.19 4.51 3.76 12.36 2.02 3.54 2.58","where_in_source":"Table 4 (FLEURS), column de","fact_url":"https://yapapp.cc/facts/voxtral-2025-gpt-4o-mini-transcribe-fleurs-de-3-76","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Soniox v5 Async Soniox","value":3.8,"low":null,"high":null,"as_written":"3.8","unit":"percent","system":"Soniox v5 Async Soniox","kind":"Cloud service","benchmark":"AA-WER","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Language is not named on the leaderboard page; the AA-WER v2.0 article describes a normalizer built on Whisper's English normalizer.","claim":"On Artificial Analysis's speech-to-text leaderboard, Soniox v5 Async Soniox had an AA-WER of 3.8 percent, its word error rate on the index's three test datasets.","quote":"Soniox v5 Async Soniox 3.8% 43.5 1.66 Details","where_in_source":"Summary of Key Metrics table","fact_url":"https://yapapp.cc/facts/aa-2026-soniox-v5-async-soniox-aa-wer","source_title":"Speech to Text (ASR) Providers Leaderboard & Comparison","source_publisher":"Artificial Analysis","source_year":"2026","source_url":"https://artificialanalysis.ai/speech-to-text","source_doi":null,"checked":"2026-10-19"},{"row":"OtterAI/speech-v3.1","value":3.8844444444444446,"low":null,"high":null,"as_written":"3.88","unit":"percent","system":"OtterAI/speech-v3.1","kind":"Cloud service","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: Proprietary): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, OtterAI/speech-v3.1 had an average word error rate of 3.88 percent.","quote":"OtterAI/speech-v3.1,3.8844444444444446","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-otterai-speech-v3-1-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"Inworld STT 1 Inworld","value":3.9,"low":null,"high":null,"as_written":"3.9","unit":"percent","system":"Inworld STT 1 Inworld","kind":"Cloud service","benchmark":"AA-WER","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Language is not named on the leaderboard page; the AA-WER v2.0 article describes a normalizer built on Whisper's English normalizer.","claim":"On Artificial Analysis's speech-to-text leaderboard, Inworld STT 1 Inworld had an AA-WER of 3.9 percent, its word error rate on the index's three test datasets.","quote":"Inworld STT 1 Inworld 3.9% 147.1 2.50 Details","where_in_source":"Summary of Key Metrics table","fact_url":"https://yapapp.cc/facts/aa-2026-inworld-stt-1-inworld-aa-wer","source_title":"Speech to Text (ASR) Providers Leaderboard & Comparison","source_publisher":"Artificial Analysis","source_year":"2026","source_url":"https://artificialanalysis.ai/speech-to-text","source_doi":null,"checked":"2026-10-19"},{"row":"GPT-4o Transcribe OpenAI","value":4,"low":null,"high":null,"as_written":"4","unit":"percent","system":"GPT-4o Transcribe OpenAI","kind":"Cloud service","benchmark":"AA-WER","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Language is not named on the leaderboard page; the AA-WER v2.0 article describes a normalizer built on Whisper's English normalizer.","claim":"On Artificial Analysis's speech-to-text leaderboard, GPT-4o Transcribe OpenAI had an AA-WER of 4.0 percent, its word error rate on the index's three test datasets.","quote":"GPT-4o Transcribe OpenAI 4.0% 37.9 6.00 Details","where_in_source":"Summary of Key Metrics table","fact_url":"https://yapapp.cc/facts/aa-2026-gpt-4o-transcribe-openai-aa-wer","source_title":"Speech to Text (ASR) Providers Leaderboard & Comparison","source_publisher":"Artificial Analysis","source_year":"2026","source_url":"https://artificialanalysis.ai/speech-to-text","source_doi":null,"checked":"2026-10-19"},{"row":"Gemini 2.5 Flash, SPGI English","value":4,"low":null,"high":null,"as_written":"4","unit":"percent","system":"Gemini 2.5 Flash","kind":"Cloud service","benchmark":"SPGI Speech","language":"English","condition":"Specialist vocabulary","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, Gemini 2.5 Flash had a word error rate of 4.00 percent on SPGI Speech.","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... Gemini 2.5 Flash 2.97 6.15 10.99 7.84 9.57 14.79 4.00 8.09 10.80","where_in_source":"Table 3 (short-form English), SPGI Speech column","fact_url":"https://yapapp.cc/facts/voxtral-2025-gemini-2-5-flash-spgi-speech-4-00","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"reson8/resonant-1-flash","value":4.02375,"low":null,"high":null,"as_written":"4.02","unit":"percent","system":"reson8/resonant-1-flash","kind":"Cloud service","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: Proprietary): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, reson8/resonant-1-flash had an average word error rate of 4.02 percent.","quote":"reson8/resonant-1-flash,4.02375","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-reson8-resonant-1-flash-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"Amazon Transcribe Amazon Bedrock","value":4.1,"low":null,"high":null,"as_written":"4.1","unit":"percent","system":"Amazon Transcribe Amazon Bedrock","kind":"Cloud service","benchmark":"AA-WER","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Language is not named on the leaderboard page; the AA-WER v2.0 article describes a normalizer built on Whisper's English normalizer.","claim":"On Artificial Analysis's speech-to-text leaderboard, Amazon Transcribe Amazon Bedrock had an AA-WER of 4.1 percent, its word error rate on the index's three test datasets.","quote":"Amazon Transcribe Amazon Bedrock 4.1% 31.3 6.00 Details","where_in_source":"Summary of Key Metrics table","fact_url":"https://yapapp.cc/facts/aa-2026-amazon-transcribe-amazon-bedrock-aa-wer","source_title":"Speech to Text (ASR) Providers Leaderboard & Comparison","source_publisher":"Artificial Analysis","source_year":"2026","source_url":"https://artificialanalysis.ai/speech-to-text","source_doi":null,"checked":"2026-10-19"},{"row":"zoom/scribe_v1","value":4.15,"low":null,"high":null,"as_written":"4.15","unit":"percent","system":"zoom/scribe_v1","kind":"Cloud service","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: Proprietary): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, zoom/scribe_v1 had an average word error rate of 4.15 percent.","quote":"zoom/scribe_v1,4.15","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-zoom-scribe-v1-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"Modulate STT Batch English VFast","value":4.2,"low":null,"high":null,"as_written":"4.2","unit":"percent","system":"Modulate STT Batch English VFast Modulate","kind":"Cloud service","benchmark":"AA-WER","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Language is not named on the leaderboard page; the AA-WER v2.0 article describes a normalizer built on Whisper's English normalizer.","claim":"On Artificial Analysis's speech-to-text leaderboard, Modulate STT Batch English VFast Modulate had an AA-WER of 4.2 percent, its word error rate on the index's three test datasets.","quote":"Modulate STT Batch English VFast Modulate 4.2% 47.8 0.42 Details","where_in_source":"Summary of Key Metrics table","fact_url":"https://yapapp.cc/facts/aa-2026-modulate-stt-batch-english-vfast-modulate-aa-wer","source_title":"Speech to Text (ASR) Providers Leaderboard & Comparison","source_publisher":"Artificial Analysis","source_year":"2026","source_url":"https://artificialanalysis.ai/speech-to-text","source_doi":null,"checked":"2026-10-19"},{"row":"Gemini 2.5 Flash, FLEURS Portuguese","value":4.23,"low":null,"high":null,"as_written":"4.23","unit":"percent","system":"Gemini 2.5 Flash","kind":"Cloud service","benchmark":"FLEURS","language":"Portuguese","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, Gemini 2.5 Flash had a word error rate of 4.23 percent on FLEURS Portuguese.","quote":"Model ar nl en fr de hi it pt es ... Gemini 2.5 Flash 25.25 6.20 4.64 6.17 4.74 6.76 2.21 4.23 3.17","where_in_source":"Table 4 (FLEURS), column pt","fact_url":"https://yapapp.cc/facts/voxtral-2025-gemini-2-5-flash-fleurs-pt-4-23","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"microsoft/azure-speech-06-2026","value":4.25875,"low":null,"high":null,"as_written":"4.26","unit":"percent","system":"microsoft/azure-speech-06-2026","kind":"Cloud service","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: Proprietary): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, microsoft/azure-speech-06-2026 had an average word error rate of 4.26 percent.","quote":"microsoft/azure-speech-06-2026,4.25875","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-microsoft-azure-speech-06-2026-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"GPT-4o mini Transcribe, MLS Spanish","value":4.28,"low":null,"high":null,"as_written":"4.28","unit":"percent","system":"GPT-4o mini Transcribe","kind":"Cloud service","benchmark":"Multilingual LibriSpeech","language":"Spanish","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 6, GPT-4o mini Transcribe had a word error rate of 4.28 percent on Multilingual LibriSpeech Spanish.","quote":"Model nl fr de it pt es ... GPT-4o mini Transcribe 8.62 4.77 5.44 10.68 5.67 4.28","where_in_source":"Table 6 (Multilingual LibriSpeech), column es","fact_url":"https://yapapp.cc/facts/voxtral-2025-gpt-4o-mini-transcribe-multilingual-librispeech-es-4-28","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Canary Qwen 2.5B, NVIDIA Replicate","value":4.3,"low":null,"high":null,"as_written":"4.3","unit":"percent","system":"Canary Qwen 2.5B, NVIDIA Replicate","kind":"Cloud service","benchmark":"AA-WER","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Language is not named on the leaderboard page; the AA-WER v2.0 article describes a normalizer built on Whisper's English normalizer.","claim":"On Artificial Analysis's speech-to-text leaderboard, Canary Qwen 2.5B, NVIDIA Replicate had an AA-WER of 4.3 percent, its word error rate on the index's three test datasets.","quote":"Canary Qwen 2.5B, NVIDIA Replicate 4.3% 8.8 0.74 Details","where_in_source":"Summary of Key Metrics table","fact_url":"https://yapapp.cc/facts/aa-2026-canary-qwen-2-5b-nvidia-replicate-aa-wer","source_title":"Speech to Text (ASR) Providers Leaderboard & Comparison","source_publisher":"Artificial Analysis","source_year":"2026","source_url":"https://artificialanalysis.ai/speech-to-text","source_doi":null,"checked":"2026-10-19"},{"row":"Gemini 2.5 Flash, MLS Spanish","value":4.39,"low":null,"high":null,"as_written":"4.39","unit":"percent","system":"Gemini 2.5 Flash","kind":"Cloud service","benchmark":"Multilingual LibriSpeech","language":"Spanish","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 6, Gemini 2.5 Flash had a word error rate of 4.39 percent on Multilingual LibriSpeech Spanish.","quote":"Model nl fr de it pt es ... Gemini 2.5 Flash 8.33 6.82 6.52 10.97 7.14 4.39","where_in_source":"Table 6 (Multilingual LibriSpeech), column es","fact_url":"https://yapapp.cc/facts/voxtral-2025-gemini-2-5-flash-multilingual-librispeech-es-4-39","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Kuhn et al. 2024, Amazon","value":4.4,"low":null,"high":null,"as_written":"4.4","unit":"percent","system":"Amazon","kind":"Cloud service","benchmark":"University lecture recordings","language":"English","condition":null,"year":2024,"conversion":null,"claim":"In the 2024 study, Amazon had an average word error rate of 4.4 percent on the English lecture recordings.","quote":"Table 4. Average Word Error Rate in percent by service provider and dataset Vendor English LibriSpeech ESL German ... Amazon 4.4 6.1 7.1 18.0","where_in_source":"Table 4","fact_url":"https://yapapp.cc/facts/kuhn-2024-amazon-english-average-wer-4-4","source_title":"Measuring the Accuracy of Automatic Speech Recognition Solutions (HTML version)","source_publisher":"arXiv","source_year":"2024","source_url":"https://arxiv.org/html/2408.16287v1","source_doi":null,"checked":"2026-10-17"},{"row":"GPT-4o Mini Transcribe OpenAI","value":4.5,"low":null,"high":null,"as_written":"4.5","unit":"percent","system":"GPT-4o Mini Transcribe OpenAI","kind":"Cloud service","benchmark":"AA-WER","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Language is not named on the leaderboard page; the AA-WER v2.0 article describes a normalizer built on Whisper's English normalizer.","claim":"On Artificial Analysis's speech-to-text leaderboard, GPT-4o Mini Transcribe OpenAI had an AA-WER of 4.5 percent, its word error rate on the index's three test datasets.","quote":"GPT-4o Mini Transcribe OpenAI 4.5% 52.3 3.00 Details","where_in_source":"Summary of Key Metrics table","fact_url":"https://yapapp.cc/facts/aa-2026-gpt-4o-mini-transcribe-openai-aa-wer","source_title":"Speech to Text (ASR) Providers Leaderboard & Comparison","source_publisher":"Artificial Analysis","source_year":"2026","source_url":"https://artificialanalysis.ai/speech-to-text","source_doi":null,"checked":"2026-10-19"},{"row":"ElevenLabs Scribe, FLEURS Portuguese","value":4.5,"low":null,"high":null,"as_written":"4.5","unit":"percent","system":"ElevenLabs Scribe","kind":"Cloud service","benchmark":"FLEURS","language":"Portuguese","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, ElevenLabs Scribe had a word error rate of 4.50 percent on FLEURS Portuguese.","quote":"Model ar nl en fr de hi it pt es ... Scribe 11.58 4.63 3.29 5.07 4.78 5.67 1.48 4.50 3.13","where_in_source":"Table 4 (FLEURS), column pt","fact_url":"https://yapapp.cc/facts/voxtral-2025-elevenlabs-scribe-fleurs-pt-4-50","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"GPT-4o mini Transcribe, FLEURS French","value":4.51,"low":null,"high":null,"as_written":"4.51","unit":"percent","system":"GPT-4o mini Transcribe","kind":"Cloud service","benchmark":"FLEURS","language":"French","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, GPT-4o mini Transcribe had a word error rate of 4.51 percent on FLEURS French.","quote":"Model ar nl en fr de hi it pt es ... GPT-4o mini Transcribe 14.02 5.54 3.19 4.51 3.76 12.36 2.02 3.54 2.58","where_in_source":"Table 4 (FLEURS), column fr","fact_url":"https://yapapp.cc/facts/voxtral-2025-gpt-4o-mini-transcribe-fleurs-fr-4-51","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"transcribe-03-2026 Cohere","value":4.6,"low":null,"high":null,"as_written":"4.6","unit":"percent","system":"transcribe-03-2026 Cohere","kind":"Cloud service","benchmark":"AA-WER","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Language is not named on the leaderboard page; the AA-WER v2.0 article describes a normalizer built on Whisper's English normalizer.","claim":"On Artificial Analysis's speech-to-text leaderboard, transcribe-03-2026 Cohere had an AA-WER of 4.6 percent, its word error rate on the index's three test datasets.","quote":"transcribe-03-2026 Cohere 4.6% 117.2 0.00 Details","where_in_source":"Summary of Key Metrics table","fact_url":"https://yapapp.cc/facts/aa-2026-transcribe-03-2026-cohere-aa-wer","source_title":"Speech to Text (ASR) Providers Leaderboard & Comparison","source_publisher":"Artificial Analysis","source_year":"2026","source_url":"https://artificialanalysis.ai/speech-to-text","source_doi":null,"checked":"2026-10-19"},{"row":"sophea/asr-k1 (preview)","value":4.614444444444445,"low":null,"high":null,"as_written":"4.61","unit":"percent","system":"sophea/asr-k1 (preview)","kind":"Cloud service","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: Proprietary): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, sophea/asr-k1 (preview) had an average word error rate of 4.61 percent.","quote":"sophea/asr-k1 (preview),4.614444444444445","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-sophea-asr-k1-preview-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"ElevenLabs Scribe, FLEURS Dutch","value":4.63,"low":null,"high":null,"as_written":"4.63","unit":"percent","system":"ElevenLabs Scribe","kind":"Cloud service","benchmark":"FLEURS","language":"Dutch","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, ElevenLabs Scribe had a word error rate of 4.63 percent on FLEURS Dutch.","quote":"Model ar nl en fr de hi it pt es ... Scribe 11.58 4.63 3.29 5.07 4.78 5.67 1.48 4.50 3.13","where_in_source":"Table 4 (FLEURS), column nl","fact_url":"https://yapapp.cc/facts/voxtral-2025-elevenlabs-scribe-fleurs-nl-4-63","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Gemini 2.5 Flash, FLEURS English","value":4.64,"low":null,"high":null,"as_written":"4.64","unit":"percent","system":"Gemini 2.5 Flash","kind":"Cloud service","benchmark":"FLEURS","language":"English","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, Gemini 2.5 Flash had a word error rate of 4.64 percent on FLEURS English.","quote":"Model ar nl en fr de hi it pt es ... Gemini 2.5 Flash 25.25 6.20 4.64 6.17 4.74 6.76 2.21 4.23 3.17","where_in_source":"Table 4 (FLEURS), column en","fact_url":"https://yapapp.cc/facts/voxtral-2025-gemini-2-5-flash-fleurs-en-4-64","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"StepAudio 2.5 ASR, StepFun StepFun","value":4.7,"low":null,"high":null,"as_written":"4.7","unit":"percent","system":"StepAudio 2.5 ASR, StepFun StepFun","kind":"Cloud service","benchmark":"AA-WER","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Language is not named on the leaderboard page; the AA-WER v2.0 article describes a normalizer built on Whisper's English normalizer.","claim":"On Artificial Analysis's speech-to-text leaderboard, StepAudio 2.5 ASR, StepFun StepFun had an AA-WER of 4.7 percent, its word error rate on the index's three test datasets.","quote":"StepAudio 2.5 ASR, StepFun StepFun 4.7% 122.4 0.37 Details","where_in_source":"Summary of Key Metrics table","fact_url":"https://yapapp.cc/facts/aa-2026-stepaudio-2-5-asr-stepfun-stepfun-aa-wer","source_title":"Speech to Text (ASR) Providers Leaderboard & Comparison","source_publisher":"Artificial Analysis","source_year":"2026","source_url":"https://artificialanalysis.ai/speech-to-text","source_doi":null,"checked":"2026-10-19"},{"row":"GPT-4o mini Transcribe, LS test-other","value":4.7,"low":null,"high":null,"as_written":"4.7","unit":"percent","system":"GPT-4o mini Transcribe","kind":"Cloud service","benchmark":"LibriSpeech test-other","language":"English","condition":"Mixed benchmark","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, GPT-4o mini Transcribe had a word error rate of 4.70 percent on LibriSpeech test-other.","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... GPT4o mini Transcribe 1.92 4.70 14.80 7.34 17.31 11.35 4.51 10.09 12.27","where_in_source":"Table 3 (short-form English), LibriSpeech test-other column","fact_url":"https://yapapp.cc/facts/voxtral-2025-gpt-4o-mini-transcribe-librispeech-test-other-4-70","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Gemini 2.5 Flash, FLEURS German","value":4.74,"low":null,"high":null,"as_written":"4.74","unit":"percent","system":"Gemini 2.5 Flash","kind":"Cloud service","benchmark":"FLEURS","language":"German","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, Gemini 2.5 Flash had a word error rate of 4.74 percent on FLEURS German.","quote":"Model ar nl en fr de hi it pt es ... Gemini 2.5 Flash 25.25 6.20 4.64 6.17 4.74 6.76 2.21 4.23 3.17","where_in_source":"Table 4 (FLEURS), column de","fact_url":"https://yapapp.cc/facts/voxtral-2025-gemini-2-5-flash-fleurs-de-4-74","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"zoom/scribe_v2_pro","value":4.768571428571428,"low":null,"high":null,"as_written":"4.77","unit":"percent","system":"zoom/scribe_v2_pro","kind":"Cloud service","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: Proprietary): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, zoom/scribe_v2_pro had an average word error rate of 4.77 percent.","quote":"zoom/scribe_v2_pro,4.768571428571428","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-zoom-scribe-v2-pro-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"GPT-4o mini Transcribe, MLS French","value":4.77,"low":null,"high":null,"as_written":"4.77","unit":"percent","system":"GPT-4o mini Transcribe","kind":"Cloud service","benchmark":"Multilingual LibriSpeech","language":"French","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 6, GPT-4o mini Transcribe had a word error rate of 4.77 percent on Multilingual LibriSpeech French.","quote":"Model nl fr de it pt es ... GPT-4o mini Transcribe 8.62 4.77 5.44 10.68 5.67 4.28","where_in_source":"Table 6 (Multilingual LibriSpeech), column fr","fact_url":"https://yapapp.cc/facts/voxtral-2025-gpt-4o-mini-transcribe-multilingual-librispeech-fr-4-77","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"ElevenLabs Scribe, FLEURS German","value":4.78,"low":null,"high":null,"as_written":"4.78","unit":"percent","system":"ElevenLabs Scribe","kind":"Cloud service","benchmark":"FLEURS","language":"German","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, ElevenLabs Scribe had a word error rate of 4.78 percent on FLEURS German.","quote":"Model ar nl en fr de hi it pt es ... Scribe 11.58 4.63 3.29 5.07 4.78 5.67 1.48 4.50 3.13","where_in_source":"Table 4 (FLEURS), column de","fact_url":"https://yapapp.cc/facts/voxtral-2025-elevenlabs-scribe-fleurs-de-4-78","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Nova-3, Telnyx Telnyx","value":4.8,"low":null,"high":null,"as_written":"4.8","unit":"percent","system":"Nova-3, Telnyx Telnyx","kind":"Cloud service","benchmark":"AA-WER","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Language is not named on the leaderboard page; the AA-WER v2.0 article describes a normalizer built on Whisper's English normalizer.","claim":"On Artificial Analysis's speech-to-text leaderboard, Nova-3, Telnyx Telnyx had an AA-WER of 4.8 percent, its word error rate on the index's three test datasets.","quote":"Nova-3, Telnyx Telnyx 4.8% 336.1 7.40 Details","where_in_source":"Summary of Key Metrics table","fact_url":"https://yapapp.cc/facts/aa-2026-nova-3-telnyx-telnyx-aa-wer","source_title":"Speech to Text (ASR) Providers Leaderboard & Comparison","source_publisher":"Artificial Analysis","source_year":"2026","source_url":"https://artificialanalysis.ai/speech-to-text","source_doi":null,"checked":"2026-10-19"},{"row":"Nova 2 Pro Amazon Bedrock","value":4.9,"low":null,"high":null,"as_written":"4.9","unit":"percent","system":"Nova 2 Pro Amazon Bedrock","kind":"Cloud service","benchmark":"AA-WER","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Language is not named on the leaderboard page; the AA-WER v2.0 article describes a normalizer built on Whisper's English normalizer.","claim":"On Artificial Analysis's speech-to-text leaderboard, Nova 2 Pro Amazon Bedrock had an AA-WER of 4.9 percent, its word error rate on the index's three test datasets.","quote":"Nova 2 Pro Amazon Bedrock 4.9% 23.8 3.10 Details","where_in_source":"Summary of Key Metrics table","fact_url":"https://yapapp.cc/facts/aa-2026-nova-2-pro-amazon-bedrock-aa-wer","source_title":"Speech to Text (ASR) Providers Leaderboard & Comparison","source_publisher":"Artificial Analysis","source_year":"2026","source_url":"https://artificialanalysis.ai/speech-to-text","source_doi":null,"checked":"2026-10-19"},{"row":"ElevenLabs Scribe, FLEURS French","value":5.07,"low":null,"high":null,"as_written":"5.07","unit":"percent","system":"ElevenLabs Scribe","kind":"Cloud service","benchmark":"FLEURS","language":"French","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, ElevenLabs Scribe had a word error rate of 5.07 percent on FLEURS French.","quote":"Model ar nl en fr de hi it pt es ... Scribe 11.58 4.63 3.29 5.07 4.78 5.67 1.48 4.50 3.13","where_in_source":"Table 4 (FLEURS), column fr","fact_url":"https://yapapp.cc/facts/voxtral-2025-elevenlabs-scribe-fleurs-fr-5-07","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Gemini 2.5 Flash Google","value":5.1,"low":null,"high":null,"as_written":"5.1","unit":"percent","system":"Gemini 2.5 Flash Google","kind":"Cloud service","benchmark":"AA-WER","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Language is not named on the leaderboard page; the AA-WER v2.0 article describes a normalizer built on Whisper's English normalizer.","claim":"On Artificial Analysis's speech-to-text leaderboard, Gemini 2.5 Flash Google had an AA-WER of 5.1 percent, its word error rate on the index's three test datasets.","quote":"Gemini 2.5 Flash Google 5.1% 71.5 6.66 Details","where_in_source":"Summary of Key Metrics table","fact_url":"https://yapapp.cc/facts/aa-2026-gemini-2-5-flash-google-aa-wer","source_title":"Speech to Text (ASR) Providers Leaderboard & Comparison","source_publisher":"Artificial Analysis","source_year":"2026","source_url":"https://artificialanalysis.ai/speech-to-text","source_doi":null,"checked":"2026-10-19"},{"row":"Gemini 2.5 Flash Lite Google","value":5.2,"low":null,"high":null,"as_written":"5.2","unit":"percent","system":"Gemini 2.5 Flash Lite Google","kind":"Cloud service","benchmark":"AA-WER","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Language is not named on the leaderboard page; the AA-WER v2.0 article describes a normalizer built on Whisper's English normalizer.","claim":"On Artificial Analysis's speech-to-text leaderboard, Gemini 2.5 Flash Lite Google had an AA-WER of 5.2 percent, its word error rate on the index's three test datasets.","quote":"Gemini 2.5 Flash Lite Google 5.2% 76.7 6.56 Details","where_in_source":"Summary of Key Metrics table","fact_url":"https://yapapp.cc/facts/aa-2026-gemini-2-5-flash-lite-google-aa-wer","source_title":"Speech to Text (ASR) Providers Leaderboard & Comparison","source_publisher":"Artificial Analysis","source_year":"2026","source_url":"https://artificialanalysis.ai/speech-to-text","source_doi":null,"checked":"2026-10-19"},{"row":"microsoft/azure-speech-07-2026","value":5.207142857142856,"low":null,"high":null,"as_written":"5.21","unit":"percent","system":"microsoft/azure-speech-07-2026","kind":"Cloud service","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: Proprietary): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, microsoft/azure-speech-07-2026 had an average word error rate of 5.21 percent.","quote":"microsoft/azure-speech-07-2026,5.207142857142856","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-microsoft-azure-speech-07-2026-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"elevenlabs/scribe_v2","value":5.365714285714285,"low":null,"high":null,"as_written":"5.37","unit":"percent","system":"elevenlabs/scribe_v2","kind":"Cloud service","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: Proprietary): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, elevenlabs/scribe_v2 had an average word error rate of 5.37 percent.","quote":"elevenlabs/scribe_v2,5.365714285714285","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-elevenlabs-scribe-v2-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet RNNT 1.1B Replicate","value":5.4,"low":null,"high":null,"as_written":"5.4","unit":"percent","system":"Parakeet RNNT 1.1B Replicate","kind":"Cloud service","benchmark":"AA-WER","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Language is not named on the leaderboard page; the AA-WER v2.0 article describes a normalizer built on Whisper's English normalizer.","claim":"On Artificial Analysis's speech-to-text leaderboard, Parakeet RNNT 1.1B Replicate had an AA-WER of 5.4 percent, its word error rate on the index's three test datasets.","quote":"Parakeet RNNT 1.1B Replicate 5.4% 6.4 1.91 Details","where_in_source":"Summary of Key Metrics table","fact_url":"https://yapapp.cc/facts/aa-2026-parakeet-rnnt-1-1b-replicate-aa-wer","source_title":"Speech to Text (ASR) Providers Leaderboard & Comparison","source_publisher":"Artificial Analysis","source_year":"2026","source_url":"https://artificialanalysis.ai/speech-to-text","source_doi":null,"checked":"2026-10-19"},{"row":"GPT-4o mini Transcribe, MLS German","value":5.44,"low":null,"high":null,"as_written":"5.44","unit":"percent","system":"GPT-4o mini Transcribe","kind":"Cloud service","benchmark":"Multilingual LibriSpeech","language":"German","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 6, GPT-4o mini Transcribe had a word error rate of 5.44 percent on Multilingual LibriSpeech German.","quote":"Model nl fr de it pt es ... GPT-4o mini Transcribe 8.62 4.77 5.44 10.68 5.67 4.28","where_in_source":"Table 6 (Multilingual LibriSpeech), column de","fact_url":"https://yapapp.cc/facts/voxtral-2025-gpt-4o-mini-transcribe-multilingual-librispeech-de-5-44","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"modulate/multilingual","value":5.447142857142857,"low":null,"high":null,"as_written":"5.45","unit":"percent","system":"modulate/multilingual","kind":"Cloud service","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: Proprietary): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, modulate/multilingual had an average word error rate of 5.45 percent.","quote":"modulate/multilingual,5.447142857142857","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-modulate-multilingual-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"ElevenLabs Scribe, CV Portuguese","value":5.46,"low":null,"high":null,"as_written":"5.46","unit":"percent","system":"ElevenLabs Scribe","kind":"Cloud service","benchmark":"Common Voice","language":"Portuguese","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, ElevenLabs Scribe had a word error rate of 5.46 percent on Common Voice Portuguese.","quote":"Model ar nl en fr de hi it pt es ... Scribe 47.03 2.38 6.59 5.44 3.52 17.29 2.99 5.46 3.27","where_in_source":"Table 5 (Common Voice), column pt","fact_url":"https://yapapp.cc/facts/voxtral-2025-elevenlabs-scribe-common-voice-pt-5-46","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Zoom Scribe v1","value":5.47,"low":null,"high":null,"as_written":"5.47","unit":"percent","system":"Zoom Scribe v1","kind":"Cloud service","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2025,"conversion":"Kind is inferred: the table marks the system as not open and the source says it is proprietary and cannot run on your own hardware.","claim":"On the same English short-form tests, Zoom Scribe v1 had an average word error rate of 5.47 percent, and the paper's table marks it as not open.","quote":"Zoom Scribe v1 No 5.47 - - - 1 IBM Granite Speech 4.0 1B","where_in_source":"Table 3, short-form English","fact_url":"https://yapapp.cc/facts/open-asr-zoom-scribe-v1-5-47","source_title":"Open ASR Leaderboard: Towards Reproducible and Transparent Multilingual and Long-Form Speech Recognition Evaluation","source_publisher":"arXiv","source_year":"2025","source_url":"https://arxiv.org/html/2510.06961v4","source_doi":null,"checked":"2026-10-17"},{"row":"Whisper Large v3 Turbo, Telnyx Telnyx","value":5.5,"low":null,"high":null,"as_written":"5.5","unit":"percent","system":"Whisper Large v3 Turbo, Telnyx Telnyx","kind":"Cloud service","benchmark":"AA-WER","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Language is not named on the leaderboard page; the AA-WER v2.0 article describes a normalizer built on Whisper's English normalizer.","claim":"On Artificial Analysis's speech-to-text leaderboard, Whisper Large v3 Turbo, Telnyx Telnyx had an AA-WER of 5.5 percent, its word error rate on the index's three test datasets.","quote":"Whisper Large v3 Turbo, Telnyx Telnyx 5.5% 37.8 15.00 Details","where_in_source":"Summary of Key Metrics table","fact_url":"https://yapapp.cc/facts/aa-2026-whisper-large-v3-turbo-telnyx-telnyx-aa-wer","source_title":"Speech to Text (ASR) Providers Leaderboard & Comparison","source_publisher":"Artificial Analysis","source_year":"2026","source_url":"https://artificialanalysis.ai/speech-to-text","source_doi":null,"checked":"2026-10-19"},{"row":"reson8/resonant-1","value":5.531428571428571,"low":null,"high":null,"as_written":"5.53","unit":"percent","system":"reson8/resonant-1","kind":"Cloud service","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: Proprietary): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, reson8/resonant-1 had an average word error rate of 5.53 percent.","quote":"reson8/resonant-1,5.531428571428571","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-reson8-resonant-1-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"GPT-4o mini Transcribe, FLEURS Dutch","value":5.54,"low":null,"high":null,"as_written":"5.54","unit":"percent","system":"GPT-4o mini Transcribe","kind":"Cloud service","benchmark":"FLEURS","language":"Dutch","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, GPT-4o mini Transcribe had a word error rate of 5.54 percent on FLEURS Dutch.","quote":"Model ar nl en fr de hi it pt es ... GPT-4o mini Transcribe 14.02 5.54 3.19 4.51 3.76 12.36 2.02 3.54 2.58","where_in_source":"Table 4 (FLEURS), column nl","fact_url":"https://yapapp.cc/facts/voxtral-2025-gpt-4o-mini-transcribe-fleurs-nl-5-54","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"ElevenLabs Scribe, FLEURS Hindi","value":5.67,"low":null,"high":null,"as_written":"5.67","unit":"percent","system":"ElevenLabs Scribe","kind":"Cloud service","benchmark":"FLEURS","language":"Hindi","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, ElevenLabs Scribe had a word error rate of 5.67 percent on FLEURS Hindi.","quote":"Model ar nl en fr de hi it pt es ... Scribe 11.58 4.63 3.29 5.07 4.78 5.67 1.48 4.50 3.13","where_in_source":"Table 4 (FLEURS), column hi","fact_url":"https://yapapp.cc/facts/voxtral-2025-elevenlabs-scribe-fleurs-hi-5-67","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Incredibly Fast Whisper Replicate","value":5.7,"low":null,"high":null,"as_written":"5.7","unit":"percent","system":"Incredibly Fast Whisper Replicate large-v3","kind":"Cloud service","benchmark":"AA-WER","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Language is not named on the leaderboard page; the AA-WER v2.0 article describes a normalizer built on Whisper's English normalizer.","claim":"On Artificial Analysis's speech-to-text leaderboard, Incredibly Fast Whisper Replicate large-v3 had an AA-WER of 5.7 percent, its word error rate on the index's three test datasets.","quote":"Incredibly Fast Whisper Replicate large-v3 5.7% 54.1 1.49 Details","where_in_source":"Summary of Key Metrics table","fact_url":"https://yapapp.cc/facts/aa-2026-incredibly-fast-whisper-replicate-large-v3-aa-wer","source_title":"Speech to Text (ASR) Providers Leaderboard & Comparison","source_publisher":"Artificial Analysis","source_year":"2026","source_url":"https://artificialanalysis.ai/speech-to-text","source_doi":null,"checked":"2026-10-19"},{"row":"Gemini 2.5 Flash, CV Dutch","value":5.73,"low":null,"high":null,"as_written":"5.73","unit":"percent","system":"Gemini 2.5 Flash","kind":"Cloud service","benchmark":"Common Voice","language":"Dutch","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, Gemini 2.5 Flash had a word error rate of 5.73 percent on Common Voice Dutch.","quote":"Model ar nl en fr de hi it pt es ... Gemini 2.5 Flash 53.62 5.73 12.31 11.86 7.25 11.09 6.64 9.42 5.88","where_in_source":"Table 5 (Common Voice), column nl","fact_url":"https://yapapp.cc/facts/voxtral-2025-gemini-2-5-flash-common-voice-nl-5-73","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"assemblyai/universal-3-5-pro","value":5.8,"low":null,"high":null,"as_written":"5.8","unit":"percent","system":"assemblyai/universal-3-5-pro","kind":"Cloud service","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: Proprietary): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, assemblyai/universal-3-5-pro had an average word error rate of 5.80 percent.","quote":"assemblyai/universal-3-5-pro,5.8","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-assemblyai-universal-3-5-pro-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"ElevenLabs Scribe, MLS French","value":5.8,"low":null,"high":null,"as_written":"5.8","unit":"percent","system":"ElevenLabs Scribe","kind":"Cloud service","benchmark":"Multilingual LibriSpeech","language":"French","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 6, ElevenLabs Scribe had a word error rate of 5.80 percent on Multilingual LibriSpeech French.","quote":"Model nl fr de it pt es ... Scribe 38.01 5.80 9.81 12.38 15.40 8.97","where_in_source":"Table 6 (Multilingual LibriSpeech), column fr","fact_url":"https://yapapp.cc/facts/voxtral-2025-elevenlabs-scribe-multilingual-librispeech-fr-5-80","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Gemini 2.5 Flash, CV Spanish","value":5.88,"low":null,"high":null,"as_written":"5.88","unit":"percent","system":"Gemini 2.5 Flash","kind":"Cloud service","benchmark":"Common Voice","language":"Spanish","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, Gemini 2.5 Flash had a word error rate of 5.88 percent on Common Voice Spanish.","quote":"Model ar nl en fr de hi it pt es ... Gemini 2.5 Flash 53.62 5.73 12.31 11.86 7.25 11.09 6.64 9.42 5.88","where_in_source":"Table 5 (Common Voice), column es","fact_url":"https://yapapp.cc/facts/voxtral-2025-gemini-2-5-flash-common-voice-es-5-88","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Rev AI Rev AI","value":5.9,"low":null,"high":null,"as_written":"5.9","unit":"percent","system":"Rev AI Rev AI","kind":"Cloud service","benchmark":"AA-WER","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Language is not named on the leaderboard page; the AA-WER v2.0 article describes a normalizer built on Whisper's English normalizer.","claim":"On Artificial Analysis's speech-to-text leaderboard, Rev AI Rev AI had an AA-WER of 5.9 percent, its word error rate on the index's three test datasets.","quote":"Rev AI Rev AI 5.9% 13.1 3.33 Details","where_in_source":"Summary of Key Metrics table","fact_url":"https://yapapp.cc/facts/aa-2026-rev-ai-rev-ai-aa-wer","source_title":"Speech to Text (ASR) Providers Leaderboard & Comparison","source_publisher":"Artificial Analysis","source_year":"2026","source_url":"https://artificialanalysis.ai/speech-to-text","source_doi":null,"checked":"2026-10-19"},{"row":"smallestai/pulse","value":5.964285714285715,"low":null,"high":null,"as_written":"5.96","unit":"percent","system":"smallestai/pulse","kind":"Cloud service","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: Proprietary): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, smallestai/pulse had an average word error rate of 5.96 percent.","quote":"smallestai/pulse,5.964285714285715","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-smallestai-pulse-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"GPT-4o mini Transcribe, CV Spanish","value":6.1,"low":null,"high":null,"as_written":"6.1","unit":"percent","system":"GPT-4o mini Transcribe","kind":"Cloud service","benchmark":"Common Voice","language":"Spanish","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, GPT-4o mini Transcribe had a word error rate of 6.10 percent on Common Voice Spanish.","quote":"Model ar nl en fr de hi it pt es ... GPT-4o mini Transcribe 51.68 6.47 13.15 10.75 6.72 39.06 6.16 9.95 6.10","where_in_source":"Table 5 (Common Voice), column es","fact_url":"https://yapapp.cc/facts/voxtral-2025-gpt-4o-mini-transcribe-common-voice-es-6-10","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Gemini 2.5 Flash, LS test-other English","value":6.15,"low":null,"high":null,"as_written":"6.15","unit":"percent","system":"Gemini 2.5 Flash","kind":"Cloud service","benchmark":"LibriSpeech test-other","language":"English","condition":"Mixed benchmark","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, Gemini 2.5 Flash had a word error rate of 6.15 percent on LibriSpeech test-other.","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... Gemini 2.5 Flash 2.97 6.15 10.99 7.84 9.57 14.79 4.00 8.09 10.80","where_in_source":"Table 3 (short-form English), LibriSpeech test-other column","fact_url":"https://yapapp.cc/facts/voxtral-2025-gemini-2-5-flash-librispeech-test-other-6-15","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"GPT-4o mini Transcribe, CV Italian","value":6.16,"low":null,"high":null,"as_written":"6.16","unit":"percent","system":"GPT-4o mini Transcribe","kind":"Cloud service","benchmark":"Common Voice","language":"Italian","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, GPT-4o mini Transcribe had a word error rate of 6.16 percent on Common Voice Italian.","quote":"Model ar nl en fr de hi it pt es ... GPT-4o mini Transcribe 51.68 6.47 13.15 10.75 6.72 39.06 6.16 9.95 6.10","where_in_source":"Table 5 (Common Voice), column it","fact_url":"https://yapapp.cc/facts/voxtral-2025-gpt-4o-mini-transcribe-common-voice-it-6-16","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Gemini 2.5 Flash, FLEURS French","value":6.17,"low":null,"high":null,"as_written":"6.17","unit":"percent","system":"Gemini 2.5 Flash","kind":"Cloud service","benchmark":"FLEURS","language":"French","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, Gemini 2.5 Flash had a word error rate of 6.17 percent on FLEURS French.","quote":"Model ar nl en fr de hi it pt es ... Gemini 2.5 Flash 25.25 6.20 4.64 6.17 4.74 6.76 2.21 4.23 3.17","where_in_source":"Table 4 (FLEURS), column fr","fact_url":"https://yapapp.cc/facts/voxtral-2025-gemini-2-5-flash-fleurs-fr-6-17","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Gemini 2.5 Flash, FLEURS Dutch","value":6.2,"low":null,"high":null,"as_written":"6.2","unit":"percent","system":"Gemini 2.5 Flash","kind":"Cloud service","benchmark":"FLEURS","language":"Dutch","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, Gemini 2.5 Flash had a word error rate of 6.20 percent on FLEURS Dutch.","quote":"Model ar nl en fr de hi it pt es ... Gemini 2.5 Flash 25.25 6.20 4.64 6.17 4.74 6.76 2.21 4.23 3.17","where_in_source":"Table 4 (FLEURS), column nl","fact_url":"https://yapapp.cc/facts/voxtral-2025-gemini-2-5-flash-fleurs-nl-6-20","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"sprag/symphony","value":6.325714285714285,"low":null,"high":null,"as_written":"6.33","unit":"percent","system":"sprag/symphony","kind":"Cloud service","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: Proprietary): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, sprag/symphony had an average word error rate of 6.33 percent.","quote":"sprag/symphony,6.325714285714285","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-sprag-symphony-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"openai/gpt-transcribe","value":6.384285714285714,"low":null,"high":null,"as_written":"6.38","unit":"percent","system":"openai/gpt-transcribe","kind":"Cloud service","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: Proprietary): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, openai/gpt-transcribe had an average word error rate of 6.38 percent.","quote":"openai/gpt-transcribe,6.384285714285714","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-openai-gpt-transcribe-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"Parakeet TDT 0.6B V2, NVIDIA NVIDIA","value":6.4,"low":null,"high":null,"as_written":"6.4","unit":"percent","system":"Parakeet TDT 0.6B V2, NVIDIA NVIDIA","kind":"Cloud service","benchmark":"AA-WER","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Language is not named on the leaderboard page; the AA-WER v2.0 article describes a normalizer built on Whisper's English normalizer.","claim":"On Artificial Analysis's speech-to-text leaderboard, Parakeet TDT 0.6B V2, NVIDIA NVIDIA had an AA-WER of 6.4 percent, its word error rate on the index's three test datasets.","quote":"Parakeet TDT 0.6B V2, NVIDIA NVIDIA 6.4% 95.2 0.00 Details","where_in_source":"Summary of Key Metrics table","fact_url":"https://yapapp.cc/facts/aa-2026-parakeet-tdt-0-6b-v2-nvidia-nvidia-aa-wer","source_title":"Speech to Text (ASR) Providers Leaderboard & Comparison","source_publisher":"Artificial Analysis","source_year":"2026","source_url":"https://artificialanalysis.ai/speech-to-text","source_doi":null,"checked":"2026-10-19"},{"row":"GPT-4o mini Transcribe, CV Dutch","value":6.47,"low":null,"high":null,"as_written":"6.47","unit":"percent","system":"GPT-4o mini Transcribe","kind":"Cloud service","benchmark":"Common Voice","language":"Dutch","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, GPT-4o mini Transcribe had a word error rate of 6.47 percent on Common Voice Dutch.","quote":"Model ar nl en fr de hi it pt es ... GPT-4o mini Transcribe 51.68 6.47 13.15 10.75 6.72 39.06 6.16 9.95 6.10","where_in_source":"Table 5 (Common Voice), column nl","fact_url":"https://yapapp.cc/facts/voxtral-2025-gpt-4o-mini-transcribe-common-voice-nl-6-47","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"meta/muse-voice-transcribe","value":6.482857142857143,"low":null,"high":null,"as_written":"6.48","unit":"percent","system":"meta/muse-voice-transcribe","kind":"Cloud service","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: Proprietary): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, meta/muse-voice-transcribe had an average word error rate of 6.48 percent.","quote":"meta/muse-voice-transcribe,6.482857142857143","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-meta-muse-voice-transcribe-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"Gemini 2.5 Flash, MLS German","value":6.52,"low":null,"high":null,"as_written":"6.52","unit":"percent","system":"Gemini 2.5 Flash","kind":"Cloud service","benchmark":"Multilingual LibriSpeech","language":"German","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 6, Gemini 2.5 Flash had a word error rate of 6.52 percent on Multilingual LibriSpeech German.","quote":"Model nl fr de it pt es ... Gemini 2.5 Flash 8.33 6.82 6.52 10.97 7.14 4.39","where_in_source":"Table 6 (Multilingual LibriSpeech), column de","fact_url":"https://yapapp.cc/facts/voxtral-2025-gemini-2-5-flash-multilingual-librispeech-de-6-52","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"ElevenLabs Scribe, CV English","value":6.59,"low":null,"high":null,"as_written":"6.59","unit":"percent","system":"ElevenLabs Scribe","kind":"Cloud service","benchmark":"Common Voice","language":"English","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, ElevenLabs Scribe had a word error rate of 6.59 percent on Common Voice English.","quote":"Model ar nl en fr de hi it pt es ... Scribe 47.03 2.38 6.59 5.44 3.52 17.29 2.99 5.46 3.27","where_in_source":"Table 5 (Common Voice), column en","fact_url":"https://yapapp.cc/facts/voxtral-2025-elevenlabs-scribe-common-voice-en-6-59","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Gemini 2.5 Flash, CV Italian","value":6.64,"low":null,"high":null,"as_written":"6.64","unit":"percent","system":"Gemini 2.5 Flash","kind":"Cloud service","benchmark":"Common Voice","language":"Italian","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, Gemini 2.5 Flash had a word error rate of 6.64 percent on Common Voice Italian.","quote":"Model ar nl en fr de hi it pt es ... Gemini 2.5 Flash 53.62 5.73 12.31 11.86 7.25 11.09 6.64 9.42 5.88","where_in_source":"Table 5 (Common Voice), column it","fact_url":"https://yapapp.cc/facts/voxtral-2025-gemini-2-5-flash-common-voice-it-6-64","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"GPT-4o mini Transcribe, CV German","value":6.72,"low":null,"high":null,"as_written":"6.72","unit":"percent","system":"GPT-4o mini Transcribe","kind":"Cloud service","benchmark":"Common Voice","language":"German","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, GPT-4o mini Transcribe had a word error rate of 6.72 percent on Common Voice German.","quote":"Model ar nl en fr de hi it pt es ... GPT-4o mini Transcribe 51.68 6.47 13.15 10.75 6.72 39.06 6.16 9.95 6.10","where_in_source":"Table 5 (Common Voice), column de","fact_url":"https://yapapp.cc/facts/voxtral-2025-gpt-4o-mini-transcribe-common-voice-de-6-72","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Gemini 2.5 Flash, FLEURS Hindi","value":6.76,"low":null,"high":null,"as_written":"6.76","unit":"percent","system":"Gemini 2.5 Flash","kind":"Cloud service","benchmark":"FLEURS","language":"Hindi","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the same Voxtral table, Gemini 2.5 Flash had a word error rate of 6.76 percent in Hindi.","quote":"Model ar nl en fr de hi it pt es Whisper large-v3 15.44 5.87 4.00 5.55 5.46 28.87 2.71 3.90 2.81 GPT-4o mini Transcribe 14.02 5.54 3.19 4.51 3.76 12.36 2.02 3.54 2.58 Gemini 2.5 Flash 25.25 6.20 4.64 6.17 4.74 6.76","where_in_source":"Table 4 (FLEURS), column hi","fact_url":"https://yapapp.cc/facts/voxtral-2025-gemini-2-5-flash-hindi-6-76","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-17"},{"row":"ElevenLabs (Pietron et al. 2026)","value":6.77,"low":null,"high":null,"as_written":"6.77","unit":"percent","system":"ElevenLabs","kind":"Cloud service","benchmark":"LibriSpeech","language":"English","condition":null,"year":2026,"conversion":"Model version not stated in the fact. Language is taken from the test set name: LibriSpeech is English.","claim":"In a 2026 preprint, the reported word error rate was 6.77 percent for ElevenLabs and 5.05 percent for Whisper large, on the paper's common benchmark table.","quote":"ElevenLabs 6.77% Whisper large 5.05%","where_in_source":"Section 5, Results on common benchmarks","fact_url":"https://yapapp.cc/facts/polish-2026-elevenlabs-6-77-whisper-5-05","source_title":"Quality of Automatic Speech Recognition: Polish Language case study, from Wav2Vec to Scribe ElevenLabs","source_publisher":"Pietron et al.","source_year":"2026","source_url":"https://arxiv.org/html/2603.02246","source_doi":null,"checked":"2026-10-17"},{"row":"Gradium Speech-to-Text Gradium","value":6.8,"low":null,"high":null,"as_written":"6.8","unit":"percent","system":"Gradium Speech-to-Text Gradium","kind":"Cloud service","benchmark":"AA-WER","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Language is not named on the leaderboard page; the AA-WER v2.0 article describes a normalizer built on Whisper's English normalizer.","claim":"On Artificial Analysis's speech-to-text leaderboard, Gradium Speech-to-Text Gradium had an AA-WER of 6.8 percent, its word error rate on the index's three test datasets.","quote":"Gradium Speech-to-Text Gradium 6.8% 2.3 13.00 Details","where_in_source":"Summary of Key Metrics table","fact_url":"https://yapapp.cc/facts/aa-2026-gradium-speech-to-text-gradium-aa-wer","source_title":"Speech to Text (ASR) Providers Leaderboard & Comparison","source_publisher":"Artificial Analysis","source_year":"2026","source_url":"https://artificialanalysis.ai/speech-to-text","source_doi":null,"checked":"2026-10-19"},{"row":"Gemini 2.5 Flash, MLS French","value":6.82,"low":null,"high":null,"as_written":"6.82","unit":"percent","system":"Gemini 2.5 Flash","kind":"Cloud service","benchmark":"Multilingual LibriSpeech","language":"French","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 6, Gemini 2.5 Flash had a word error rate of 6.82 percent on Multilingual LibriSpeech French.","quote":"Model nl fr de it pt es ... Gemini 2.5 Flash 8.33 6.82 6.52 10.97 7.14 4.39","where_in_source":"Table 6 (Multilingual LibriSpeech), column fr","fact_url":"https://yapapp.cc/facts/voxtral-2025-gemini-2-5-flash-multilingual-librispeech-fr-6-82","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"ElevenLabs Scribe, VoxPopuli English","value":6.95,"low":null,"high":null,"as_written":"6.95","unit":"percent","system":"ElevenLabs Scribe","kind":"Cloud service","benchmark":"VoxPopuli","language":"English","condition":"Mixed benchmark","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, ElevenLabs Scribe had a word error rate of 6.95 percent on VoxPopuli.","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... ElevenLabs Scribe 1.80 3.44 10.52 6.95 10.62 8.35 3.16 7.39 9.16","where_in_source":"Table 3 (short-form English), VoxPopuli column","fact_url":"https://yapapp.cc/facts/voxtral-2025-elevenlabs-scribe-voxpopuli-6-95","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Gemini 2.5 Flash, MLS Portuguese","value":7.14,"low":null,"high":null,"as_written":"7.14","unit":"percent","system":"Gemini 2.5 Flash","kind":"Cloud service","benchmark":"Multilingual LibriSpeech","language":"Portuguese","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 6, Gemini 2.5 Flash had a word error rate of 7.14 percent on Multilingual LibriSpeech Portuguese.","quote":"Model nl fr de it pt es ... Gemini 2.5 Flash 8.33 6.82 6.52 10.97 7.14 4.39","where_in_source":"Table 6 (Multilingual LibriSpeech), column pt","fact_url":"https://yapapp.cc/facts/voxtral-2025-gemini-2-5-flash-multilingual-librispeech-pt-7-14","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Gemini 2.5 Flash, CV German","value":7.25,"low":null,"high":null,"as_written":"7.25","unit":"percent","system":"Gemini 2.5 Flash","kind":"Cloud service","benchmark":"Common Voice","language":"German","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, Gemini 2.5 Flash had a word error rate of 7.25 percent on Common Voice German.","quote":"Model ar nl en fr de hi it pt es ... Gemini 2.5 Flash 53.62 5.73 12.31 11.86 7.25 11.09 6.64 9.42 5.88","where_in_source":"Table 5 (Common Voice), column de","fact_url":"https://yapapp.cc/facts/voxtral-2025-gemini-2-5-flash-common-voice-de-7-25","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"GPT-4o mini Transcribe, VoxPopuli","value":7.34,"low":null,"high":null,"as_written":"7.34","unit":"percent","system":"GPT-4o mini Transcribe","kind":"Cloud service","benchmark":"VoxPopuli","language":"English","condition":"Mixed benchmark","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, GPT-4o mini Transcribe had a word error rate of 7.34 percent on VoxPopuli.","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... GPT4o mini Transcribe 1.92 4.70 14.80 7.34 17.31 11.35 4.51 10.09 12.27","where_in_source":"Table 3 (short-form English), VoxPopuli column","fact_url":"https://yapapp.cc/facts/voxtral-2025-gpt-4o-mini-transcribe-voxpopuli-7-34","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"ElevenLabs Scribe, Earnings-21 English","value":7.39,"low":null,"high":null,"as_written":"7.39","unit":"percent","system":"ElevenLabs Scribe","kind":"Cloud service","benchmark":"Earnings-21 (10-minute segments)","language":"English","condition":"Specialist vocabulary","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, ElevenLabs Scribe had a word error rate of 7.39 percent on Earnings-21 (10-minute segments).","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... ElevenLabs Scribe 1.80 3.44 10.52 6.95 10.62 8.35 3.16 7.39 9.16","where_in_source":"Table 3 (long-form English), Earnings-21 (10-minute segments) column","fact_url":"https://yapapp.cc/facts/voxtral-2025-elevenlabs-scribe-earnings-21-10-minute-segments-7-39","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Gemini 2.5 Flash, VoxPopuli English","value":7.84,"low":null,"high":null,"as_written":"7.84","unit":"percent","system":"Gemini 2.5 Flash","kind":"Cloud service","benchmark":"VoxPopuli","language":"English","condition":"Mixed benchmark","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, Gemini 2.5 Flash had a word error rate of 7.84 percent on VoxPopuli.","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... Gemini 2.5 Flash 2.97 6.15 10.99 7.84 9.57 14.79 4.00 8.09 10.80","where_in_source":"Table 3 (short-form English), VoxPopuli column","fact_url":"https://yapapp.cc/facts/voxtral-2025-gemini-2-5-flash-voxpopuli-7-84","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Gemini 2.5 Flash, EN short-form avg","value":8.04,"low":null,"high":null,"as_written":"8.04","unit":"percent","system":"Gemini 2.5 Flash","kind":"Cloud service","benchmark":"English short-form average (Voxtral paper)","language":"English","condition":"Mixed benchmark","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 2, Gemini 2.5 Flash had a macro-average word error rate of 8.04 percent across the English short-form test sets.","quote":"Gemini 2.5 Flash 88.64 ± 2.28 4.54 ± 0.07 8.04","where_in_source":"Table 2, En Short WER column","fact_url":"https://yapapp.cc/facts/voxtral-2025-gemini-2-5-flash-en-short-form-average-8-04","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Gemini 2.5 Flash, Earnings-21 English","value":8.09,"low":null,"high":null,"as_written":"8.09","unit":"percent","system":"Gemini 2.5 Flash","kind":"Cloud service","benchmark":"Earnings-21 (10-minute segments)","language":"English","condition":"Specialist vocabulary","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, Gemini 2.5 Flash had a word error rate of 8.09 percent on Earnings-21 (10-minute segments).","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... Gemini 2.5 Flash 2.97 6.15 10.99 7.84 9.57 14.79 4.00 8.09 10.80","where_in_source":"Table 3 (long-form English), Earnings-21 (10-minute segments) column","fact_url":"https://yapapp.cc/facts/voxtral-2025-gemini-2-5-flash-earnings-21-10-minute-segments-8-09","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Kuhn et al. 2024, Deepgram","value":8.3,"low":null,"high":null,"as_written":"8.3","unit":"percent","system":"Deepgram","kind":"Cloud service","benchmark":"University lecture recordings","language":"English","condition":null,"year":2024,"conversion":null,"claim":"In the 2024 study, Deepgram had an average word error rate of 8.3 percent on the English lecture recordings.","quote":"Table 4. Average Word Error Rate in percent by service provider and dataset Vendor English LibriSpeech ESL German ... Deepgram 8.3 12.8 11.4 19.3","where_in_source":"Table 4","fact_url":"https://yapapp.cc/facts/kuhn-2024-deepgram-english-average-wer-8-3","source_title":"Measuring the Accuracy of Automatic Speech Recognition Solutions (HTML version)","source_publisher":"arXiv","source_year":"2024","source_url":"https://arxiv.org/html/2408.16287v1","source_doi":null,"checked":"2026-10-17"},{"row":"Gemini 2.5 Flash, MLS Dutch","value":8.33,"low":null,"high":null,"as_written":"8.33","unit":"percent","system":"Gemini 2.5 Flash","kind":"Cloud service","benchmark":"Multilingual LibriSpeech","language":"Dutch","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 6, Gemini 2.5 Flash had a word error rate of 8.33 percent on Multilingual LibriSpeech Dutch.","quote":"Model nl fr de it pt es ... Gemini 2.5 Flash 8.33 6.82 6.52 10.97 7.14 4.39","where_in_source":"Table 6 (Multilingual LibriSpeech), column nl","fact_url":"https://yapapp.cc/facts/voxtral-2025-gemini-2-5-flash-multilingual-librispeech-nl-8-33","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"ElevenLabs Scribe, CHiME-4 English","value":8.35,"low":null,"high":null,"as_written":"8.35","unit":"percent","system":"ElevenLabs Scribe","kind":"Cloud service","benchmark":"CHiME-4","language":"English","condition":"Noisy or distant","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, ElevenLabs Scribe had a word error rate of 8.35 percent on CHiME-4.","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... ElevenLabs Scribe 1.80 3.44 10.52 6.95 10.62 8.35 3.16 7.39 9.16","where_in_source":"Table 3 (short-form English), CHiME-4 column","fact_url":"https://yapapp.cc/facts/voxtral-2025-elevenlabs-scribe-chime-4-8-35","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"GPT-4o mini Transcribe, MLS Dutch","value":8.62,"low":null,"high":null,"as_written":"8.62","unit":"percent","system":"GPT-4o mini Transcribe","kind":"Cloud service","benchmark":"Multilingual LibriSpeech","language":"Dutch","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 6, GPT-4o mini Transcribe had a word error rate of 8.62 percent on Multilingual LibriSpeech Dutch.","quote":"Model nl fr de it pt es ... GPT-4o mini Transcribe 8.62 4.77 5.44 10.68 5.67 4.28","where_in_source":"Table 6 (Multilingual LibriSpeech), column nl","fact_url":"https://yapapp.cc/facts/voxtral-2025-gpt-4o-mini-transcribe-multilingual-librispeech-nl-8-62","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"ElevenLabs Scribe, MLS Spanish","value":8.97,"low":null,"high":null,"as_written":"8.97","unit":"percent","system":"ElevenLabs Scribe","kind":"Cloud service","benchmark":"Multilingual LibriSpeech","language":"Spanish","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 6, ElevenLabs Scribe had a word error rate of 8.97 percent on Multilingual LibriSpeech Spanish.","quote":"Model nl fr de it pt es ... Scribe 38.01 5.80 9.81 12.38 15.40 8.97","where_in_source":"Table 6 (Multilingual LibriSpeech), column es","fact_url":"https://yapapp.cc/facts/voxtral-2025-elevenlabs-scribe-multilingual-librispeech-es-8-97","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"ElevenLabs Scribe, Earnings-22 English","value":9.16,"low":null,"high":null,"as_written":"9.16","unit":"percent","system":"ElevenLabs Scribe","kind":"Cloud service","benchmark":"Earnings-22 (10-minute segments)","language":"English","condition":"Specialist vocabulary","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, ElevenLabs Scribe had a word error rate of 9.16 percent on Earnings-22 (10-minute segments).","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... ElevenLabs Scribe 1.80 3.44 10.52 6.95 10.62 8.35 3.16 7.39 9.16","where_in_source":"Table 3 (long-form English), Earnings-22 (10-minute segments) column","fact_url":"https://yapapp.cc/facts/voxtral-2025-elevenlabs-scribe-earnings-22-10-minute-segments-9-16","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Gemini 2.5 Flash, CV Portuguese","value":9.42,"low":null,"high":null,"as_written":"9.42","unit":"percent","system":"Gemini 2.5 Flash","kind":"Cloud service","benchmark":"Common Voice","language":"Portuguese","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, Gemini 2.5 Flash had a word error rate of 9.42 percent on Common Voice Portuguese.","quote":"Model ar nl en fr de hi it pt es ... Gemini 2.5 Flash 53.62 5.73 12.31 11.86 7.25 11.09 6.64 9.42 5.88","where_in_source":"Table 5 (Common Voice), column pt","fact_url":"https://yapapp.cc/facts/voxtral-2025-gemini-2-5-flash-common-voice-pt-9-42","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Gemini 2.5 Flash, Switchboard English","value":9.57,"low":null,"high":null,"as_written":"9.57","unit":"percent","system":"Gemini 2.5 Flash","kind":"Cloud service","benchmark":"Switchboard","language":"English","condition":"Conversational","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, Gemini 2.5 Flash had a word error rate of 9.57 percent on Switchboard.","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... Gemini 2.5 Flash 2.97 6.15 10.99 7.84 9.57 14.79 4.00 8.09 10.80","where_in_source":"Table 3 (short-form English), Switchboard column","fact_url":"https://yapapp.cc/facts/voxtral-2025-gemini-2-5-flash-switchboard-9-57","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"ElevenLabs Scribe, MLS German","value":9.81,"low":null,"high":null,"as_written":"9.81","unit":"percent","system":"ElevenLabs Scribe","kind":"Cloud service","benchmark":"Multilingual LibriSpeech","language":"German","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 6, ElevenLabs Scribe had a word error rate of 9.81 percent on Multilingual LibriSpeech German.","quote":"Model nl fr de it pt es ... Scribe 38.01 5.80 9.81 12.38 15.40 8.97","where_in_source":"Table 6 (Multilingual LibriSpeech), column de","fact_url":"https://yapapp.cc/facts/voxtral-2025-elevenlabs-scribe-multilingual-librispeech-de-9-81","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"GPT-4o mini Transcribe, CV Portuguese","value":9.95,"low":null,"high":null,"as_written":"9.95","unit":"percent","system":"GPT-4o mini Transcribe","kind":"Cloud service","benchmark":"Common Voice","language":"Portuguese","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, GPT-4o mini Transcribe had a word error rate of 9.95 percent on Common Voice Portuguese.","quote":"Model ar nl en fr de hi it pt es ... GPT-4o mini Transcribe 51.68 6.47 13.15 10.75 6.72 39.06 6.16 9.95 6.10","where_in_source":"Table 5 (Common Voice), column pt","fact_url":"https://yapapp.cc/facts/voxtral-2025-gpt-4o-mini-transcribe-common-voice-pt-9-95","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"GPT-4o mini Transcribe, Earnings-21","value":10.09,"low":null,"high":null,"as_written":"10.09","unit":"percent","system":"GPT-4o mini Transcribe","kind":"Cloud service","benchmark":"Earnings-21 (10-minute segments)","language":"English","condition":"Specialist vocabulary","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, GPT-4o mini Transcribe had a word error rate of 10.09 percent on Earnings-21 (10-minute segments).","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... GPT4o mini Transcribe 1.92 4.70 14.80 7.34 17.31 11.35 4.51 10.09 12.27","where_in_source":"Table 3 (long-form English), Earnings-21 (10-minute segments) column","fact_url":"https://yapapp.cc/facts/voxtral-2025-gpt-4o-mini-transcribe-earnings-21-10-minute-segments-10-09","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Whisper Large v3 Replicate large-v3","value":10.1,"low":null,"high":null,"as_written":"10.1","unit":"percent","system":"Whisper Large v3 Replicate large-v3","kind":"Cloud service","benchmark":"AA-WER","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Language is not named on the leaderboard page; the AA-WER v2.0 article describes a normalizer built on Whisper's English normalizer.","claim":"On Artificial Analysis's speech-to-text leaderboard, Whisper Large v3 Replicate large-v3 had an AA-WER of 10.1 percent, its word error rate on the index's three test datasets.","quote":"Whisper Large v3 Replicate large-v3 10.1% 2.7 4.23 Details","where_in_source":"Summary of Key Metrics table","fact_url":"https://yapapp.cc/facts/aa-2026-whisper-large-v3-replicate-large-v3-aa-wer","source_title":"Speech to Text (ASR) Providers Leaderboard & Comparison","source_publisher":"Artificial Analysis","source_year":"2026","source_url":"https://artificialanalysis.ai/speech-to-text","source_doi":null,"checked":"2026-10-19"},{"row":"ElevenLabs Scribe, GigaSpeech English","value":10.52,"low":null,"high":null,"as_written":"10.52","unit":"percent","system":"ElevenLabs Scribe","kind":"Cloud service","benchmark":"GigaSpeech","language":"English","condition":"Mixed benchmark","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, ElevenLabs Scribe had a word error rate of 10.52 percent on GigaSpeech.","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... ElevenLabs Scribe 1.80 3.44 10.52 6.95 10.62 8.35 3.16 7.39 9.16","where_in_source":"Table 3 (short-form English), GigaSpeech column","fact_url":"https://yapapp.cc/facts/voxtral-2025-elevenlabs-scribe-gigaspeech-10-52","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"ElevenLabs Scribe, Switchboard English","value":10.62,"low":null,"high":null,"as_written":"10.62","unit":"percent","system":"ElevenLabs Scribe","kind":"Cloud service","benchmark":"Switchboard","language":"English","condition":"Conversational","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, ElevenLabs Scribe had a word error rate of 10.62 percent on Switchboard.","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... ElevenLabs Scribe 1.80 3.44 10.52 6.95 10.62 8.35 3.16 7.39 9.16","where_in_source":"Table 3 (short-form English), Switchboard column","fact_url":"https://yapapp.cc/facts/voxtral-2025-elevenlabs-scribe-switchboard-10-62","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"GPT-4o mini Transcribe, MLS Italian","value":10.68,"low":null,"high":null,"as_written":"10.68","unit":"percent","system":"GPT-4o mini Transcribe","kind":"Cloud service","benchmark":"Multilingual LibriSpeech","language":"Italian","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 6, GPT-4o mini Transcribe had a word error rate of 10.68 percent on Multilingual LibriSpeech Italian.","quote":"Model nl fr de it pt es ... GPT-4o mini Transcribe 8.62 4.77 5.44 10.68 5.67 4.28","where_in_source":"Table 6 (Multilingual LibriSpeech), column it","fact_url":"https://yapapp.cc/facts/voxtral-2025-gpt-4o-mini-transcribe-multilingual-librispeech-it-10-68","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"GPT-4o mini Transcribe, CV French","value":10.75,"low":null,"high":null,"as_written":"10.75","unit":"percent","system":"GPT-4o mini Transcribe","kind":"Cloud service","benchmark":"Common Voice","language":"French","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, GPT-4o mini Transcribe had a word error rate of 10.75 percent on Common Voice French.","quote":"Model ar nl en fr de hi it pt es ... GPT-4o mini Transcribe 51.68 6.47 13.15 10.75 6.72 39.06 6.16 9.95 6.10","where_in_source":"Table 5 (Common Voice), column fr","fact_url":"https://yapapp.cc/facts/voxtral-2025-gpt-4o-mini-transcribe-common-voice-fr-10-75","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Gemini 2.5 Flash, Earnings-22 English","value":10.8,"low":null,"high":null,"as_written":"10.8","unit":"percent","system":"Gemini 2.5 Flash","kind":"Cloud service","benchmark":"Earnings-22 (10-minute segments)","language":"English","condition":"Specialist vocabulary","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, Gemini 2.5 Flash had a word error rate of 10.80 percent on Earnings-22 (10-minute segments).","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... Gemini 2.5 Flash 2.97 6.15 10.99 7.84 9.57 14.79 4.00 8.09 10.80","where_in_source":"Table 3 (long-form English), Earnings-22 (10-minute segments) column","fact_url":"https://yapapp.cc/facts/voxtral-2025-gemini-2-5-flash-earnings-22-10-minute-segments-10-80","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Gemini 2.5 Flash, MLS Italian","value":10.97,"low":null,"high":null,"as_written":"10.97","unit":"percent","system":"Gemini 2.5 Flash","kind":"Cloud service","benchmark":"Multilingual LibriSpeech","language":"Italian","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 6, Gemini 2.5 Flash had a word error rate of 10.97 percent on Multilingual LibriSpeech Italian.","quote":"Model nl fr de it pt es ... Gemini 2.5 Flash 8.33 6.82 6.52 10.97 7.14 4.39","where_in_source":"Table 6 (Multilingual LibriSpeech), column it","fact_url":"https://yapapp.cc/facts/voxtral-2025-gemini-2-5-flash-multilingual-librispeech-it-10-97","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Gemini 2.5 Flash, GigaSpeech English","value":10.99,"low":null,"high":null,"as_written":"10.99","unit":"percent","system":"Gemini 2.5 Flash","kind":"Cloud service","benchmark":"GigaSpeech","language":"English","condition":"Mixed benchmark","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, Gemini 2.5 Flash had a word error rate of 10.99 percent on GigaSpeech.","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... Gemini 2.5 Flash 2.97 6.15 10.99 7.84 9.57 14.79 4.00 8.09 10.80","where_in_source":"Table 3 (short-form English), GigaSpeech column","fact_url":"https://yapapp.cc/facts/voxtral-2025-gemini-2-5-flash-gigaspeech-10-99","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Gemini 2.5 Flash, CV Hindi","value":11.09,"low":null,"high":null,"as_written":"11.09","unit":"percent","system":"Gemini 2.5 Flash","kind":"Cloud service","benchmark":"Common Voice","language":"Hindi","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, Gemini 2.5 Flash had a word error rate of 11.09 percent on Common Voice Hindi.","quote":"Model ar nl en fr de hi it pt es ... Gemini 2.5 Flash 53.62 5.73 12.31 11.86 7.25 11.09 6.64 9.42 5.88","where_in_source":"Table 5 (Common Voice), column hi","fact_url":"https://yapapp.cc/facts/voxtral-2025-gemini-2-5-flash-common-voice-hi-11-09","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Kuhn et al. 2024, IBM","value":11.2,"low":null,"high":null,"as_written":"11.2","unit":"percent","system":"IBM","kind":"Cloud service","benchmark":"University lecture recordings","language":"English","condition":null,"year":2024,"conversion":null,"claim":"In the 2024 study, IBM had an average word error rate of 11.2 percent on the English lecture recordings.","quote":"Table 4. Average Word Error Rate in percent by service provider and dataset Vendor English LibriSpeech ESL German ... IBM 11.2 13.2 17.3 20.6","where_in_source":"Table 4","fact_url":"https://yapapp.cc/facts/kuhn-2024-ibm-english-average-wer-11-2","source_title":"Measuring the Accuracy of Automatic Speech Recognition Solutions (HTML version)","source_publisher":"arXiv","source_year":"2024","source_url":"https://arxiv.org/html/2408.16287v1","source_doi":null,"checked":"2026-10-17"},{"row":"GPT-4o mini Transcribe, CHiME-4 English","value":11.35,"low":null,"high":null,"as_written":"11.35","unit":"percent","system":"GPT-4o mini Transcribe","kind":"Cloud service","benchmark":"CHiME-4","language":"English","condition":"Noisy or distant","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, GPT-4o mini Transcribe had a word error rate of 11.35 percent on CHiME-4.","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... GPT4o mini Transcribe 1.92 4.70 14.80 7.34 17.31 11.35 4.51 10.09 12.27","where_in_source":"Table 3 (short-form English), CHiME-4 column","fact_url":"https://yapapp.cc/facts/voxtral-2025-gpt-4o-mini-transcribe-chime-4-11-35","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"ElevenLabs Scribe, FLEURS Arabic","value":11.58,"low":null,"high":null,"as_written":"11.58","unit":"percent","system":"ElevenLabs Scribe","kind":"Cloud service","benchmark":"FLEURS","language":"Arabic","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 4, ElevenLabs Scribe had a word error rate of 11.58 percent on FLEURS Arabic.","quote":"Model ar nl en fr de hi it pt es ... Scribe 11.58 4.63 3.29 5.07 4.78 5.67 1.48 4.50 3.13","where_in_source":"Table 4 (FLEURS), column ar","fact_url":"https://yapapp.cc/facts/voxtral-2025-elevenlabs-scribe-fleurs-ar-11-58","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Gemini 2.5 Flash, CV French","value":11.86,"low":null,"high":null,"as_written":"11.86","unit":"percent","system":"Gemini 2.5 Flash","kind":"Cloud service","benchmark":"Common Voice","language":"French","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, Gemini 2.5 Flash had a word error rate of 11.86 percent on Common Voice French.","quote":"Model ar nl en fr de hi it pt es ... Gemini 2.5 Flash 53.62 5.73 12.31 11.86 7.25 11.09 6.64 9.42 5.88","where_in_source":"Table 5 (Common Voice), column fr","fact_url":"https://yapapp.cc/facts/voxtral-2025-gemini-2-5-flash-common-voice-fr-11-86","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"GPT-4o mini Transcribe, Earnings-22","value":12.27,"low":null,"high":null,"as_written":"12.27","unit":"percent","system":"GPT-4o mini Transcribe","kind":"Cloud service","benchmark":"Earnings-22 (10-minute segments)","language":"English","condition":"Specialist vocabulary","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, GPT-4o mini Transcribe had a word error rate of 12.27 percent on Earnings-22 (10-minute segments).","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... GPT4o mini Transcribe 1.92 4.70 14.80 7.34 17.31 11.35 4.51 10.09 12.27","where_in_source":"Table 3 (long-form English), Earnings-22 (10-minute segments) column","fact_url":"https://yapapp.cc/facts/voxtral-2025-gpt-4o-mini-transcribe-earnings-22-10-minute-segments-12-27","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Gemini 2.5 Flash, CV English","value":12.31,"low":null,"high":null,"as_written":"12.31","unit":"percent","system":"Gemini 2.5 Flash","kind":"Cloud service","benchmark":"Common Voice","language":"English","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, Gemini 2.5 Flash had a word error rate of 12.31 percent on Common Voice English.","quote":"Model ar nl en fr de hi it pt es ... Gemini 2.5 Flash 53.62 5.73 12.31 11.86 7.25 11.09 6.64 9.42 5.88","where_in_source":"Table 5 (Common Voice), column en","fact_url":"https://yapapp.cc/facts/voxtral-2025-gemini-2-5-flash-common-voice-en-12-31","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"GPT-4o mini Transcribe, FLEURS Hindi","value":12.36,"low":null,"high":null,"as_written":"12.36","unit":"percent","system":"GPT-4o mini Transcribe","kind":"Cloud service","benchmark":"FLEURS","language":"Hindi","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the same Voxtral table, GPT-4o mini Transcribe had a word error rate of 12.36 percent in Hindi.","quote":"Model ar nl en fr de hi it pt es Whisper large-v3 15.44 5.87 4.00 5.55 5.46 28.87 2.71 3.90 2.81 GPT-4o mini Transcribe 14.02 5.54 3.19 4.51 3.76 12.36 2.02 3.54 2.58 Gemini 2.5 Flash 25.25 6.20 4.64 6.17 4.74 6.76","where_in_source":"Table 4 (FLEURS), column hi","fact_url":"https://yapapp.cc/facts/voxtral-2025-gpt-4o-mini-transcribe-hindi-12-36","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-17"},{"row":"ElevenLabs Scribe, MLS Italian","value":12.38,"low":null,"high":null,"as_written":"12.38","unit":"percent","system":"ElevenLabs Scribe","kind":"Cloud service","benchmark":"Multilingual LibriSpeech","language":"Italian","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 6, ElevenLabs Scribe had a word error rate of 12.38 percent on Multilingual LibriSpeech Italian.","quote":"Model nl fr de it pt es ... Scribe 38.01 5.80 9.81 12.38 15.40 8.97","where_in_source":"Table 6 (Multilingual LibriSpeech), column it","fact_url":"https://yapapp.cc/facts/voxtral-2025-elevenlabs-scribe-multilingual-librispeech-it-12-38","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"GPT-4o mini Transcribe, CV English","value":13.15,"low":null,"high":null,"as_written":"13.15","unit":"percent","system":"GPT-4o mini Transcribe","kind":"Cloud service","benchmark":"Common Voice","language":"English","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, GPT-4o mini Transcribe had a word error rate of 13.15 percent on Common Voice English.","quote":"Model ar nl en fr de hi it pt es ... GPT-4o mini Transcribe 51.68 6.47 13.15 10.75 6.72 39.06 6.16 9.95 6.10","where_in_source":"Table 5 (Common Voice), column en","fact_url":"https://yapapp.cc/facts/voxtral-2025-gpt-4o-mini-transcribe-common-voice-en-13-15","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Qwen3.5 Omni Flash Alibaba Cloud","value":13.5,"low":null,"high":null,"as_written":"13.5","unit":"percent","system":"Qwen3.5 Omni Flash Alibaba Cloud","kind":"Cloud service","benchmark":"AA-WER","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Language is not named on the leaderboard page; the AA-WER v2.0 article describes a normalizer built on Whisper's English normalizer.","claim":"On Artificial Analysis's speech-to-text leaderboard, Qwen3.5 Omni Flash Alibaba Cloud had an AA-WER of 13.5 percent, its word error rate on the index's three test datasets.","quote":"Qwen3.5 Omni Flash Alibaba Cloud 13.5% 74.4 0.00 Details","where_in_source":"Summary of Key Metrics table","fact_url":"https://yapapp.cc/facts/aa-2026-qwen3-5-omni-flash-alibaba-cloud-aa-wer","source_title":"Speech to Text (ASR) Providers Leaderboard & Comparison","source_publisher":"Artificial Analysis","source_year":"2026","source_url":"https://artificialanalysis.ai/speech-to-text","source_doi":null,"checked":"2026-10-19"},{"row":"GPT-4o mini Transcribe, FLEURS Arabic","value":14.02,"low":null,"high":null,"as_written":"14.02","unit":"percent","system":"GPT-4o mini Transcribe","kind":"Cloud service","benchmark":"FLEURS","language":"Arabic","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the same Voxtral table, GPT-4o mini Transcribe had a word error rate of 14.02 percent in Arabic.","quote":"Model ar nl en fr de hi it pt es Whisper large-v3 15.44 5.87 4.00 5.55 5.46 28.87 2.71 3.90 2.81 GPT-4o mini Transcribe 14.02 5.54 3.19 4.51 3.76 12.36 2.02 3.54 2.58 Gemini 2.5 Flash 25.25 6.20 4.64 6.17 4.74 6.76","where_in_source":"Table 4 (FLEURS), column ar","fact_url":"https://yapapp.cc/facts/voxtral-2025-gpt-4o-mini-transcribe-arabic-14-02","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-17"},{"row":"Gemini 2.5 Flash, CHiME-4 English","value":14.79,"low":null,"high":null,"as_written":"14.79","unit":"percent","system":"Gemini 2.5 Flash","kind":"Cloud service","benchmark":"CHiME-4","language":"English","condition":"Noisy or distant","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, Gemini 2.5 Flash had a word error rate of 14.79 percent on CHiME-4.","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... Gemini 2.5 Flash 2.97 6.15 10.99 7.84 9.57 14.79 4.00 8.09 10.80","where_in_source":"Table 3 (short-form English), CHiME-4 column","fact_url":"https://yapapp.cc/facts/voxtral-2025-gemini-2-5-flash-chime-4-14-79","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"GPT-4o mini Transcribe, GigaSpeech","value":14.8,"low":null,"high":null,"as_written":"14.8","unit":"percent","system":"GPT-4o mini Transcribe","kind":"Cloud service","benchmark":"GigaSpeech","language":"English","condition":"Mixed benchmark","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, GPT-4o mini Transcribe had a word error rate of 14.80 percent on GigaSpeech.","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... GPT4o mini Transcribe 1.92 4.70 14.80 7.34 17.31 11.35 4.51 10.09 12.27","where_in_source":"Table 3 (short-form English), GigaSpeech column","fact_url":"https://yapapp.cc/facts/voxtral-2025-gpt-4o-mini-transcribe-gigaspeech-14-80","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"ElevenLabs Scribe, MLS Portuguese","value":15.4,"low":null,"high":null,"as_written":"15.4","unit":"percent","system":"ElevenLabs Scribe","kind":"Cloud service","benchmark":"Multilingual LibriSpeech","language":"Portuguese","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 6, ElevenLabs Scribe had a word error rate of 15.40 percent on Multilingual LibriSpeech Portuguese.","quote":"Model nl fr de it pt es ... Scribe 38.01 5.80 9.81 12.38 15.40 8.97","where_in_source":"Table 6 (Multilingual LibriSpeech), column pt","fact_url":"https://yapapp.cc/facts/voxtral-2025-elevenlabs-scribe-multilingual-librispeech-pt-15-40","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"ElevenLabs Scribe, CV Hindi","value":17.29,"low":null,"high":null,"as_written":"17.29","unit":"percent","system":"ElevenLabs Scribe","kind":"Cloud service","benchmark":"Common Voice","language":"Hindi","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, ElevenLabs Scribe had a word error rate of 17.29 percent on Common Voice Hindi.","quote":"Model ar nl en fr de hi it pt es ... Scribe 47.03 2.38 6.59 5.44 3.52 17.29 2.99 5.46 3.27","where_in_source":"Table 5 (Common Voice), column hi","fact_url":"https://yapapp.cc/facts/voxtral-2025-elevenlabs-scribe-common-voice-hi-17-29","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"GPT-4o mini Transcribe, Switchboard","value":17.31,"low":null,"high":null,"as_written":"17.31","unit":"percent","system":"GPT-4o mini Transcribe","kind":"Cloud service","benchmark":"Switchboard","language":"English","condition":"Conversational","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 3, GPT-4o mini Transcribe had a word error rate of 17.31 percent on Switchboard.","quote":"Model LS-C LS-O GS VP SB C-4 SPGI E21 10m E22 10m ... GPT4o mini Transcribe 1.92 4.70 14.80 7.34 17.31 11.35 4.51 10.09 12.27","where_in_source":"Table 3 (short-form English), Switchboard column","fact_url":"https://yapapp.cc/facts/voxtral-2025-gpt-4o-mini-transcribe-switchboard-17-31","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Jimenez et al. 2026, Mexican","value":18,"low":null,"high":null,"as_written":"18","unit":"percent","system":"YouTube automatic captions","kind":"Cloud service","benchmark":"Latin American Spanish dialect recordings","language":"Mexican Spanish","condition":"Accented or dialect","year":2026,"conversion":null,"claim":"In a 2026 test of YouTube's automatic Spanish captions, Mexican Spanish had an overall word error rate of 18 percent.","quote":"Mexican Spanish exhibit an overall WER value of 18%","where_in_source":"Results (Table 3 discussion), PDF body","fact_url":"https://yapapp.cc/facts/jimenez-2026-mexican-spanish-wer-18","source_title":"Dialect and Gender Bias in YouTube's Spanish Captioning System","source_publisher":"arXiv (Computation and Language, cs.CL), preprint 2602.24002","source_year":"2026","source_url":"https://arxiv.org/pdf/2602.24002","source_doi":null,"checked":"2026-10-17"},{"row":"gladia/solaria-3","value":18.898571428571426,"low":null,"high":null,"as_written":"18.9","unit":"percent","system":"gladia/solaria-3","kind":"Cloud service","benchmark":"Open ASR Leaderboard (English average)","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Kind follows the leaderboard's License column (license: Proprietary): Proprietary is Cloud service, any other license is Open model.","claim":"On the Open ASR Leaderboard's English short-form results, gladia/solaria-3 had an average word error rate of 18.90 percent.","quote":"gladia/solaria-3,18.898571428571426","where_in_source":"english_short_latest.csv, column avg","fact_url":"https://yapapp.cc/facts/hf-2026-gladia-solaria-3-avg-wer","source_title":"Open ASR Leaderboard: English short-form results (english_short_latest.csv)","source_publisher":"Hugging Face, hf-audio organization","source_year":"2026","source_url":"https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/main/english_short_latest.csv","source_doi":null,"checked":"2026-10-19"},{"row":"Kuhn et al. 2024, Google","value":20.1,"low":null,"high":null,"as_written":"20.1","unit":"percent","system":"Google","kind":"Cloud service","benchmark":"University lecture recordings","language":"English","condition":null,"year":2024,"conversion":null,"claim":"In a 2024 study of eleven speech recognition services on university lecture recordings, Google's average word error rate on English was 20.1 percent.","quote":"Google 20.1 23.6 28.1 18.1 IBM 11.2 13.2 17.3 20.6","where_in_source":"Table 4, English column","fact_url":"https://yapapp.cc/facts/kuhn-2024-google-english-average-wer-20-1","source_title":"Measuring the Accuracy of Automatic Speech Recognition Solutions","source_publisher":"arXiv (preprint)","source_year":"2024","source_url":"https://arxiv.org/pdf/2408.16287","source_doi":null,"checked":"2026-10-17"},{"row":"Jimenez et al. 2026, Argentinian","value":24,"low":null,"high":null,"as_written":"24","unit":"percent","system":"YouTube automatic captions","kind":"Cloud service","benchmark":"Latin American Spanish dialect recordings","language":"Argentinian Spanish","condition":"Accented or dialect","year":2026,"conversion":null,"claim":"In a 2026 test of YouTube's automatic Spanish captions, Argentinian Spanish had the highest overall word error rate of the dialects tested, at 24 percent.","quote":"Argentinian Spanish yields the highest overall WER at 24%","where_in_source":"Results (Table 3 discussion), PDF body","fact_url":"https://yapapp.cc/facts/jimenez-2026-argentinian-spanish-wer-24","source_title":"Dialect and Gender Bias in YouTube's Spanish Captioning System","source_publisher":"arXiv (Computation and Language, cs.CL), preprint 2602.24002","source_year":"2026","source_url":"https://arxiv.org/pdf/2602.24002","source_doi":null,"checked":"2026-10-17"},{"row":"Gemini 2.5 Flash, FLEURS Arabic","value":25.25,"low":null,"high":null,"as_written":"25.25","unit":"percent","system":"Gemini 2.5 Flash","kind":"Cloud service","benchmark":"FLEURS","language":"Arabic","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the same Voxtral table, Gemini 2.5 Flash had a word error rate of 25.25 percent in Arabic, the highest of the three systems.","quote":"Model ar nl en fr de hi it pt es Whisper large-v3 15.44 5.87 4.00 5.55 5.46 28.87 2.71 3.90 2.81 GPT-4o mini Transcribe 14.02 5.54 3.19 4.51 3.76 12.36 2.02 3.54 2.58 Gemini 2.5 Flash 25.25 6.20 4.64 6.17 4.74 6.76","where_in_source":"Table 4 (FLEURS), column ar","fact_url":"https://yapapp.cc/facts/voxtral-2025-gemini-2-5-flash-arabic-25-25","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-17"},{"row":"Chirp Google","value":31.2,"low":null,"high":null,"as_written":"31.2","unit":"percent","system":"Chirp Google","kind":"Cloud service","benchmark":"AA-WER","language":"English","condition":"Mixed benchmark","year":2026,"conversion":"Language is not named on the leaderboard page; the AA-WER v2.0 article describes a normalizer built on Whisper's English normalizer.","claim":"On Artificial Analysis's speech-to-text leaderboard, Chirp Google had an AA-WER of 31.2 percent, its word error rate on the index's three test datasets.","quote":"Chirp Google 31.2% 14.0 16.00 Details","where_in_source":"Summary of Key Metrics table","fact_url":"https://yapapp.cc/facts/aa-2026-chirp-google-aa-wer","source_title":"Speech to Text (ASR) Providers Leaderboard & Comparison","source_publisher":"Artificial Analysis","source_year":"2026","source_url":"https://artificialanalysis.ai/speech-to-text","source_doi":null,"checked":"2026-10-19"},{"row":"ElevenLabs Scribe, MLS Dutch","value":38.01,"low":null,"high":null,"as_written":"38.01","unit":"percent","system":"ElevenLabs Scribe","kind":"Cloud service","benchmark":"Multilingual LibriSpeech","language":"Dutch","condition":"Clean read speech","year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 6, ElevenLabs Scribe had a word error rate of 38.01 percent on Multilingual LibriSpeech Dutch.","quote":"Model nl fr de it pt es ... Scribe 38.01 5.80 9.81 12.38 15.40 8.97","where_in_source":"Table 6 (Multilingual LibriSpeech), column nl","fact_url":"https://yapapp.cc/facts/voxtral-2025-elevenlabs-scribe-multilingual-librispeech-nl-38-01","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"GPT-4o mini Transcribe, CV Hindi","value":39.06,"low":null,"high":null,"as_written":"39.06","unit":"percent","system":"GPT-4o mini Transcribe","kind":"Cloud service","benchmark":"Common Voice","language":"Hindi","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, GPT-4o mini Transcribe had a word error rate of 39.06 percent on Common Voice Hindi.","quote":"Model ar nl en fr de hi it pt es ... GPT-4o mini Transcribe 51.68 6.47 13.15 10.75 6.72 39.06 6.16 9.95 6.10","where_in_source":"Table 5 (Common Voice), column hi","fact_url":"https://yapapp.cc/facts/voxtral-2025-gpt-4o-mini-transcribe-common-voice-hi-39-06","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"ElevenLabs Scribe, CV Arabic","value":47.03,"low":null,"high":null,"as_written":"47.03","unit":"percent","system":"ElevenLabs Scribe","kind":"Cloud service","benchmark":"Common Voice","language":"Arabic","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, ElevenLabs Scribe had a word error rate of 47.03 percent on Common Voice Arabic.","quote":"Model ar nl en fr de hi it pt es ... Scribe 47.03 2.38 6.59 5.44 3.52 17.29 2.99 5.46 3.27","where_in_source":"Table 5 (Common Voice), column ar","fact_url":"https://yapapp.cc/facts/voxtral-2025-elevenlabs-scribe-common-voice-ar-47-03","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"GPT-4o mini Transcribe, CV Arabic","value":51.68,"low":null,"high":null,"as_written":"51.68","unit":"percent","system":"GPT-4o mini Transcribe","kind":"Cloud service","benchmark":"Common Voice","language":"Arabic","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, GPT-4o mini Transcribe had a word error rate of 51.68 percent on Common Voice Arabic.","quote":"Model ar nl en fr de hi it pt es ... GPT-4o mini Transcribe 51.68 6.47 13.15 10.75 6.72 39.06 6.16 9.95 6.10","where_in_source":"Table 5 (Common Voice), column ar","fact_url":"https://yapapp.cc/facts/voxtral-2025-gpt-4o-mini-transcribe-common-voice-ar-51-68","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Gemini 2.5 Flash, CV Arabic","value":53.62,"low":null,"high":null,"as_written":"53.62","unit":"percent","system":"Gemini 2.5 Flash","kind":"Cloud service","benchmark":"Common Voice","language":"Arabic","condition":null,"year":2025,"conversion":null,"claim":"In the Voxtral paper's Table 5, Gemini 2.5 Flash had a word error rate of 53.62 percent on Common Voice Arabic.","quote":"Model ar nl en fr de hi it pt es ... Gemini 2.5 Flash 53.62 5.73 12.31 11.86 7.25 11.09 6.64 9.42 5.88","where_in_source":"Table 5 (Common Voice), column ar","fact_url":"https://yapapp.cc/facts/voxtral-2025-gemini-2-5-flash-common-voice-ar-53-62","source_title":"Voxtral","source_publisher":"arXiv (Audio and Speech Processing)","source_year":"2025","source_url":"https://arxiv.org/html/2507.13264","source_doi":null,"checked":"2026-10-19"},{"row":"Apple SpeechTranscriber, Italian (our test)","value":4.15,"low":null,"high":null,"as_written":"4.15","unit":"percent","system":"Apple SpeechTranscriber","kind":"Built-in dictation","benchmark":"FLEURS","language":"Italian","condition":"Clean read speech","year":2026,"conversion":null,"claim":"In our October 2026 benchmark on a Mac with an M2 Pro, Apple SpeechTranscriber had a word error rate of 4.15 percent on 100 Italian recordings from the FLEURS test set.","quote":null,"where_in_source":null,"fact_url":"https://yapapp.cc/facts/m-fleurs-2026-10-apple-it-wer","source_title":null,"source_publisher":null,"source_year":null,"source_url":null,"source_doi":null,"checked":"2026-10-19"},{"row":"Apple SpeechTranscriber, Spanish (our test)","value":4.3,"low":null,"high":null,"as_written":"4.3","unit":"percent","system":"Apple SpeechTranscriber","kind":"Built-in dictation","benchmark":"FLEURS","language":"Spanish","condition":"Clean read speech","year":2026,"conversion":null,"claim":"In our October 2026 benchmark on a Mac with an M2 Pro, Apple SpeechTranscriber had a word error rate of 4.30 percent on 100 Spanish recordings from the FLEURS test set.","quote":null,"where_in_source":null,"fact_url":"https://yapapp.cc/facts/m-fleurs-2026-10-apple-es-wer","source_title":null,"source_publisher":null,"source_year":null,"source_url":null,"source_doi":null,"checked":"2026-10-19"},{"row":"Apple SpeechTranscriber, French (our test)","value":6.59,"low":null,"high":null,"as_written":"6.59","unit":"percent","system":"Apple SpeechTranscriber","kind":"Built-in dictation","benchmark":"FLEURS","language":"French","condition":"Clean read speech","year":2026,"conversion":null,"claim":"In our October 2026 benchmark on a Mac with an M2 Pro, Apple SpeechTranscriber had a word error rate of 6.59 percent on 100 French recordings from the FLEURS test set.","quote":null,"where_in_source":null,"fact_url":"https://yapapp.cc/facts/m-fleurs-2026-10-apple-fr-wer","source_title":null,"source_publisher":null,"source_year":null,"source_url":null,"source_doi":null,"checked":"2026-10-19"},{"row":"Apple SpeechTranscriber, English (our test)","value":6.73,"low":null,"high":null,"as_written":"6.73","unit":"percent","system":"Apple SpeechTranscriber","kind":"Built-in dictation","benchmark":"FLEURS","language":"English","condition":"Clean read speech","year":2026,"conversion":null,"claim":"In our October 2026 benchmark on a Mac with an M2 Pro, Apple SpeechTranscriber had a word error rate of 6.73 percent on 100 English recordings from the FLEURS test set.","quote":null,"where_in_source":null,"fact_url":"https://yapapp.cc/facts/m-fleurs-2026-10-apple-en-wer","source_title":null,"source_publisher":null,"source_year":null,"source_url":null,"source_doi":null,"checked":"2026-10-19"},{"row":"Apple SpeechTranscriber, German (our test)","value":6.87,"low":null,"high":null,"as_written":"6.87","unit":"percent","system":"Apple SpeechTranscriber","kind":"Built-in dictation","benchmark":"FLEURS","language":"German","condition":"Clean read speech","year":2026,"conversion":null,"claim":"In our October 2026 benchmark on a Mac with an M2 Pro, Apple SpeechTranscriber had a word error rate of 6.87 percent on 100 German recordings from the FLEURS test set.","quote":null,"where_in_source":null,"fact_url":"https://yapapp.cc/facts/m-fleurs-2026-10-apple-de-wer","source_title":null,"source_publisher":null,"source_year":null,"source_url":null,"source_doi":null,"checked":"2026-10-19"},{"row":"Apple SpeechTranscriber, Portuguese (our test)","value":100,"low":null,"high":null,"as_written":"100","unit":"percent","system":"Apple SpeechTranscriber","kind":"Built-in dictation","benchmark":"FLEURS","language":"Portuguese","condition":"Clean read speech","year":2026,"conversion":null,"claim":"In our October 2026 benchmark on a Mac with an M2 Pro, Apple SpeechTranscriber had a word error rate of 100.00 percent on 100 Portuguese recordings from the FLEURS test set.","quote":null,"where_in_source":null,"fact_url":"https://yapapp.cc/facts/m-fleurs-2026-10-apple-pt-wer","source_title":null,"source_publisher":null,"source_year":null,"source_url":null,"source_doi":null,"checked":"2026-10-19"},{"row":"Conformer, no LM","value":2.1,"low":null,"high":null,"as_written":"2.1","unit":"percent","system":"Conformer (no language model)","kind":"Research system","benchmark":"LibriSpeech test-clean","language":"English","condition":"Clean read speech","year":2020,"conversion":null,"claim":"The Conformer model reached word error rates of 2.1 percent on LibriSpeech test-clean and 4.3 percent on test-other without a language model, and 1.9 and 3.9 percent with one.","quote":"our model achieves WER of 2.1%/4.3% without using a language model and 1.9%/3.9% with an external language model on test/testother","where_in_source":"Abstract","fact_url":"https://yapapp.cc/facts/conformer-librispeech-2-1-4-3","source_title":"Conformer: Convolution-augmented Transformer for Speech Recognition","source_publisher":"arXiv (Interspeech 2020)","source_year":"2020","source_url":"https://arxiv.org/abs/2005.08100","source_doi":null,"checked":"2026-10-16"},{"row":"Close-talk DNN, WSJ studio","value":3.7,"low":null,"high":null,"as_written":"3.7","unit":"percent","system":"DNN acoustic model (close-talk baseline)","kind":"Research system","benchmark":"WSJ close-talking studio recordings","language":"English","condition":"Clean read speech","year":2017,"conversion":"Language is English from the Wall Street Journal (WSJ) corpus named in the source. Year is the 2017 publication year, as collection is not stated.","claim":"In the same study, close-talking read speech recorded in a studio had a word error rate of 3.7 percent with DNN models.","quote":"The Word Error Rate (WER%) obtained by decoding the close-talking WSJ sentences recorded in the recording studio is 3.7","where_in_source":"Section 4.1","fact_url":"https://yapapp.cc/facts/close-talk-baseline-3-7-percent","source_title":"Realistic multi-microphone data simulation for distant speech recognition","source_publisher":"arXiv preprint","source_year":"2017","source_url":"https://ar5iv.labs.arxiv.org/html/1711.09470","source_doi":null,"checked":"2026-10-16"},{"row":"Ravanelli et al. 2017, close-talk","value":3.7,"low":null,"high":null,"as_written":"3.7","unit":"percent","system":"DNN trained on clean WSJ data","kind":"Research system","benchmark":"Wall Street Journal, close-talking studio recording","language":"English","condition":"Clean read speech","year":2017,"conversion":"Language is taken from the test set name: the Wall Street Journal corpus is English.","claim":"Decoding close-talking Wall Street Journal sentences recorded in a studio gave a word error rate of 3.7%, using DNN models trained on clean WSJ data.","quote":"is 3.7% (using DNN models trained with the original clean WSJ data set).","where_in_source":"Section 4.1, Close-talking baseline","fact_url":"https://yapapp.cc/facts/ravanelli-2017-close-talk-wer-3-7","source_title":"Realistic multi-microphone data simulation for distant speech recognition","source_publisher":"arXiv (preprint, eess.AS)","source_year":"2017","source_url":"https://arxiv.org/pdf/1711.09470","source_doi":null,"checked":"2026-10-16"},{"row":"Saon et al. 2017, IBM, Switchboard","value":5.5,"low":null,"high":null,"as_written":"5.5","unit":"percent","system":"IBM English conversational recognizer","kind":"Research system","benchmark":"Switchboard (Hub5 2000)","language":"English","condition":"Conversational","year":2017,"conversion":null,"claim":"IBM's own English conversational telephone recognizer reached a word error rate of 5.5 percent on the Switchboard subset of the Hub5 2000 evaluation.","quote":"to the level of 5.5%/10.3% on the Switchboard/CallHome subsets of the Hub5 2000 evaluation","where_in_source":"Abstract","fact_url":"https://yapapp.cc/facts/saon-2017-machine-switchboard-5-5","source_title":"English Conversational Telephone Speech Recognition by Humans and Machines","source_publisher":"arXiv (cs.CL), IBM Research","source_year":"2017","source_url":"https://arxiv.org/pdf/1703.02136","source_doi":null,"checked":"2026-10-17"},{"row":"Xiong et al. 2017, system, Switchboard","value":5.8,"low":null,"high":null,"as_written":"5.8","unit":"percent","system":"Microsoft conversational ASR system (2017)","kind":"Research system","benchmark":"Switchboard (Hub5 2000)","language":"English","condition":"Conversational","year":2017,"conversion":"The source calls this the NIST 2000 Switchboard test, which is the Hub5 2000 set. The quote gives 5.8 and 11.0 for Switchboard and CallHome respectively.","claim":"Microsoft's automated system reached word error rates of 5.8 percent on Switchboard and 11.0 percent on CallHome, slightly below the professional transcribers.","quote":"our automated system establishes a new state of the art, and edges past the human benchmark, achieving error rates of 5.8% and 11.0%, respectively","where_in_source":"Abstract","fact_url":"https://yapapp.cc/facts/xiong-2017-system-5-8","source_title":"Achieving Human Parity in Conversational Speech Recognition","source_publisher":"arXiv (Microsoft Research)","source_year":"2017","source_url":"https://arxiv.org/abs/1610.05256","source_doi":null,"checked":"2026-10-16"},{"row":"Lin et al. 2025, Extra tags","value":5.81,"low":null,"high":null,"as_written":"5.81","unit":"percent","system":"Whisper, fine-tuned with hesitation tags (Extra)","kind":"Research system","benchmark":"SLaTE 2025 challenge test set","language":"English","condition":"Accented or dialect","year":2025,"conversion":null,"claim":"In a 2025 challenge on non-native English, a fine-tuned Whisper system scored 5.81 percent word error rate when it labeled hesitations, against 6.47 percent when it removed them.","quote":"Our challenge system achieved 6.47% WER (Pure) and 5.81% WER (Extra).","where_in_source":"Abstract","fact_url":"https://yapapp.cc/facts/lin-2025-hesitation-tags-challenge-wer","source_title":"Acoustically Precise Hesitation Tagging Is Essential for End-to-End Verbatim Transcription Systems","source_publisher":"arXiv (SLaTE 2025 workshop)","source_year":"2025","source_url":"https://arxiv.org/abs/2506.04076","source_doi":null,"checked":"2026-10-17"},{"row":"Saon et al. 2017, IBM, CallHome","value":10.3,"low":null,"high":null,"as_written":"10.3","unit":"percent","system":"IBM English conversational recognizer","kind":"Research system","benchmark":"CallHome (Hub5 2000)","language":"English","condition":"Conversational","year":2017,"conversion":null,"claim":"IBM's own English conversational telephone recognizer reached a word error rate of 10.3 percent on the CallHome subset of the Hub5 2000 evaluation.","quote":"to the level of 5.5%/10.3% on the Switchboard/CallHome subsets of the Hub5 2000 evaluation","where_in_source":"Abstract","fact_url":"https://yapapp.cc/facts/saon-2017-machine-callhome-10-3","source_title":"English Conversational Telephone Speech Recognition by Humans and Machines","source_publisher":"arXiv (cs.CL), IBM Research","source_year":"2017","source_url":"https://arxiv.org/pdf/1703.02136","source_doi":null,"checked":"2026-10-17"},{"row":"Kurmanji Whisper v3 2024","value":10.5,"low":null,"high":null,"as_written":"10.5","unit":"percent","system":"Whisper version 3, fine-tuned on Kurmanji","kind":"Research system","benchmark":"Specialized Northern Kurdish test set","language":"Northern Kurdish (Kurmanji)","condition":null,"year":2024,"conversion":null,"claim":"A 2024 study that fine-tuned Whisper version 3 on about 68 hours of Northern Kurdish (Kurmanji) speech reached a word error rate of 10.5 percent on a specialized test set, using its best fine-tuning strategy.","quote":"achieving a Word Error Rate (WER) of 10.5% and Character Error Rate (CER) of 5.7% with Whisper version 3","where_in_source":"Abstract","fact_url":"https://yapapp.cc/facts/kurmanji-whisper-v3-wer-10-5","source_title":"End-to-End Transformer-based Automatic Speech Recognition for Northern Kurdish: A Pioneering Approach","source_publisher":"arXiv","source_year":"2024","source_url":"https://arxiv.org/abs/2410.16330","source_doi":null,"checked":"2026-10-17"},{"row":"IndicWhisper, FLEURS Hindi","value":11.4,"low":null,"high":null,"as_written":"11.4","unit":"percent","system":"IndicWhisper","kind":"Research system","benchmark":"FLEURS","language":"Hindi","condition":null,"year":2023,"conversion":null,"claim":"On the FLEURS subset of the Vistaar Hindi benchmark, IndicWhisper had a word error rate of 11.4 percent.","quote":"IndicWhisper 10.3 12.0 11.4 15.0 7.6 12.0 26.8 13.6","where_in_source":"Table 3","fact_url":"https://yapapp.cc/facts/bhogale-2023-indicwhisper-hindi-fleurs-11-4","source_title":"Vistaar: Diverse Benchmarks and Training Sets for Indian Language ASR","source_publisher":"Interspeech 2023 (arXiv:2305.15386)","source_year":"2023","source_url":"https://arxiv.org/html/2305.15386","source_doi":null,"checked":"2026-10-17"},{"row":"IndicWhisper, Hindi average","value":13.6,"low":null,"high":null,"as_written":"13.6","unit":"percent","system":"IndicWhisper","kind":"Research system","benchmark":"Vistaar Hindi benchmark","language":"Hindi","condition":"Mixed benchmark","year":2023,"conversion":"Average word error rate across the seven Hindi test sets in the Vistaar benchmark.","claim":"On the Hindi subset of the Vistaar benchmark, IndicWhisper, which fine-tunes Whisper on Indian-language data, had an average word error rate of 13.6 percent across seven test sets.","quote":"IndicWhisper 10.3 12.0 11.4 15.0 7.6 12.0 26.8 13.6","where_in_source":"Table 3","fact_url":"https://yapapp.cc/facts/bhogale-2023-indicwhisper-hindi-13-6","source_title":"Vistaar: Diverse Benchmarks and Training Sets for Indian Language ASR","source_publisher":"Interspeech 2023 (arXiv:2305.15386)","source_year":"2023","source_url":"https://arxiv.org/html/2305.15386","source_doi":null,"checked":"2026-10-17"},{"row":"XLS-R fine-tuned, Swiss German","value":14,"low":null,"high":null,"as_written":"14","unit":"percent","system":"Wav2Vec2 XLS-R (fine-tuned)","kind":"Research system","benchmark":"Swiss German corpus test set","language":"Swiss German","condition":"Accented or dialect","year":2024,"conversion":"Figure reported by the corpus authors and cited by Dolev and colleagues, who did not rerun it.","claim":"The Swiss German corpus authors fine-tuned a Wav2Vec2 XLS-R model on their data and reached a word error rate of 14.0 percent on their test set, as Dolev and colleagues report.","quote":"They fine-tuned a Wav2Vec2 XLS-R model on the corpus and reached a WER of 14.0 and a BLEU score of 74.7 on their test set.","where_in_source":"Section 2, related work on Swiss German corpora","fact_url":"https://yapapp.cc/facts/dolev-2024-fine-tuned-xls-r-swiss-german-14-percent","source_title":"Does Whisper Understand Swiss German? An Automatic, Qualitative and Human Evaluation","source_publisher":"Proceedings of the 11th Workshop on NLP for Similar Languages, Varieties and Dialects (VarDial 2024)","source_year":"2024","source_url":"https://arxiv.org/html/2404.19310v2","source_doi":null,"checked":"2026-10-17"},{"row":"DNN-HMM, RT03S (2012)","value":18.5,"low":null,"high":null,"as_written":"18.5","unit":"percent","system":"DNN-HMM acoustic model","kind":"Research system","benchmark":"Switchboard RT03S","language":"English","condition":null,"year":2012,"conversion":null,"claim":"In a 2012 review, a deep neural network reduced the word error rate of a GMM-HMM baseline from 27.4 to 18.5 percent on a Switchboard telephone task.","quote":"Using a DNN reduced the word error rate (WER) from the 27.4% of the baseline GMM-HMM (trained with BMMI) to 18.5%","where_in_source":"Switchboard speech recognition task, Table 2","fact_url":"https://yapapp.cc/facts/hinton-2012-switchboard-wer-27-4-to-18-5","source_title":"Deep Neural Networks for Acoustic Modeling in Speech Recognition: The Shared Views of Four Research Groups","source_publisher":"IEEE Signal Processing Magazine","source_year":"2012","source_url":"https://www.cs.toronto.edu/~hinton/absps/DNN-2012-proof.pdf","source_doi":"10.1109/MSP.2012.2205597","checked":"2026-10-17"},{"row":"Jacobs et al. 2025, isiXhosa","value":80.4,"low":null,"high":null,"as_written":"80.4","unit":"percent","system":"Whisper medium, fine-tuned on child speech","kind":"Research system","benchmark":"Preschool oral narratives (isiXhosa)","language":"isiXhosa","condition":"Children","year":2025,"conversion":null,"claim":"After fine-tuning Whisper medium on the same amount of transcribed child speech, the word error rate was 80.4 percent for isiXhosa, against 47.4 percent for Afrikaans.","quote":"the WER for Afrikaans is significantly better than for isiXhosa: 47.4% compared to 80.4% for Whisper medium","where_in_source":"Results section, first row of Table II","fact_url":"https://yapapp.cc/facts/jacobs-2025-isixhosa-wer-vs-afrikaans","source_title":"Speech Recognition for Automatically Assessing Afrikaans and isiXhosa Preschool Oral Narratives","source_publisher":"arXiv (Stellenbosch University and University of Edinburgh)","source_year":"2025","source_url":"https://arxiv.org/pdf/2501.06478","source_doi":null,"checked":"2026-10-17"},{"row":"Saon et al. 2017, humans, Switchboard","value":5.1,"low":null,"high":null,"as_written":"5.1","unit":"percent","system":"Human transcribers","kind":"Human transcribers","benchmark":"Switchboard (Hub5 2000)","language":"English","condition":"Conversational","year":2017,"conversion":null,"claim":"In the IBM study, the best of three human transcribers reached a word error rate of 5.1 percent on the Switchboard telephone test after quality checks.","quote":"could only get down to 5.1% for the best transcriber after quality checks","where_in_source":"Section 2, human transcription experiments","fact_url":"https://yapapp.cc/facts/saon-2017-human-switchboard-5-1","source_title":"English Conversational Telephone Speech Recognition by Humans and Machines","source_publisher":"arXiv (cs.CL), IBM Research","source_year":"2017","source_url":"https://arxiv.org/pdf/1703.02136","source_doi":null,"checked":"2026-10-17"},{"row":"Xiong et al. 2017, humans, Switchboard","value":5.9,"low":null,"high":null,"as_written":"5.9","unit":"percent","system":"Human transcribers","kind":"Human transcribers","benchmark":"Switchboard (Hub5 2000)","language":"English","condition":"Conversational","year":2017,"conversion":"The source calls this the NIST 2000 Switchboard test, which is the Hub5 2000 set.","claim":"On the NIST 2000 Switchboard test of telephone conversations, professional human transcribers had a word error rate of 5.9 percent, and 11.3 percent on the harder CallHome conversations.","quote":"The error rate of professional transcribers is 5.9% for the Switchboard portion of the data, in which newly acquainted pairs of people discuss an assigned topic, and 11.3% for the CallHome portion","where_in_source":"Abstract","fact_url":"https://yapapp.cc/facts/xiong-2017-professional-transcribers-5-9","source_title":"Achieving Human Parity in Conversational Speech Recognition","source_publisher":"arXiv (Microsoft Research)","source_year":"2017","source_url":"https://arxiv.org/abs/1610.05256","source_doi":null,"checked":"2026-10-16"},{"row":"Saon et al. 2017, humans, CallHome","value":6.8,"low":null,"high":null,"as_written":"6.8","unit":"percent","system":"Human transcribers","kind":"Human transcribers","benchmark":"CallHome (Hub5 2000)","language":"English","condition":"Conversational","year":2017,"conversion":null,"claim":"In the same IBM study, the best human transcriber reached a word error rate of 6.8 percent on the CallHome telephone test.","quote":"achieved a surprisingly low 6.8% WER for CallHome","where_in_source":"Section 2, human transcription experiments","fact_url":"https://yapapp.cc/facts/saon-2017-human-callhome-6-8","source_title":"English Conversational Telephone Speech Recognition by Humans and Machines","source_publisher":"arXiv (cs.CL), IBM Research","source_year":"2017","source_url":"https://arxiv.org/pdf/1703.02136","source_doi":null,"checked":"2026-10-17"}]}