Fact ยท Speech recognition
32sources
The speech recognition word error rates dataset draws on 32 distinct sources.
Calculated from 597 data points on October 19, 2026
- How it is calculated
- Count of the distinct sources behind the dataset's 597 rows: 32.
A statistic of the dataset Speech Recognition Accuracy: Word Error Rates, calculated by code from its published rows.
Calculated from
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- On the FLEURS subset of the Vistaar Hindi benchmark, IndicWhisper had a word error rate of 11.4 percent.
- In the same study, close-talking read speech recorded in a studio had a word error rate of 3.7 percent with DNN models.
- 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.
- 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.
- 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.
- In a 2024 baseline test, the OpenAI Whisper model had a word error rate of 18.8 percent on whispered speech.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- On the Open ASR Leaderboard's English short-form results, Edge0/ARK-ASR-3B had an average word error rate of 6.53 percent.
- 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.
- On the Open ASR Leaderboard's English short-form results, elevenlabs/scribe_v2 had an average word error rate of 5.37 percent.
- 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.
- 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.
- 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.
- On the Open ASR Leaderboard's English short-form results, FermionResearch/Phonon-2 had an average word error rate of 7.23 percent.
- 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.
- On the Open ASR Leaderboard's English short-form results, gladia/solaria-3 had an average word error rate of 18.90 percent.
- 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.
- 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.
- 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.
- On the Open ASR Leaderboard's English short-form results, HojoAI/Hojo-ASR-V1 had an average word error rate of 6.07 percent.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- On the Open ASR Leaderboard's English short-form results, meta/muse-voice-transcribe had an average word error rate of 6.48 percent.
- 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.
- 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.
- 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.
- On the Open ASR Leaderboard's English short-form results, microsoft/VibeVoice-ASR-HF had an average word error rate of 8.23 percent.
- 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.
- 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.
- 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.
- On the Open ASR Leaderboard's English short-form results, modulate/multilingual had an average word error rate of 5.45 percent.
- On the Open ASR Leaderboard's English short-form results, nvidia/canary-1b-v2 had an average word error rate of 7.91 percent.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- On the Open ASR Leaderboard's English short-form results, openai/gpt-transcribe had an average word error rate of 6.38 percent.
- On the Open ASR Leaderboard's English short-form results, openai/whisper-large-v3 had an average word error rate of 7.84 percent.
- 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.
- 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.
- On the Open ASR Leaderboard's English short-form results, OtterAI/speech-v3.1 had an average word error rate of 3.88 percent.
- 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.
- 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.
- On the Open ASR Leaderboard's English short-form results, reson8/resonant-1 had an average word error rate of 5.53 percent.
- On the Open ASR Leaderboard's English short-form results, reson8/resonant-1-flash had an average word error rate of 4.02 percent.
- On the Open ASR Leaderboard's English short-form results, smallestai/pulse had an average word error rate of 5.96 percent.
- On the Open ASR Leaderboard's English short-form results, sophea/asr-k1 (preview) had an average word error rate of 4.61 percent.
- 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.
- 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.
- On the Open ASR Leaderboard's English short-form results, sprag/symphony had an average word error rate of 6.33 percent.
- 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.
- On the Open ASR Leaderboard's English short-form results, zoom/scribe_v1 had an average word error rate of 4.15 percent.
- On the Open ASR Leaderboard's English short-form results, zoom/scribe_v2_pro had an average word error rate of 4.77 percent.
- 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.
- 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.
- 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.
- In a 2026 test of YouTube's automatic Spanish captions, Mexican Spanish had an overall word error rate of 18 percent.
- In the 2024 study, Amazon had an average word error rate of 4.4 percent on the English lecture recordings.
- In the 2024 study, Deepgram had an average word error rate of 8.3 percent on the English lecture recordings.
- 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.
- In the 2024 study, IBM had an average word error rate of 11.2 percent on the English lecture recordings.
- In the same 2024 study, Whisper large-v2 had an average word error rate of 2.9 percent on the English lecture recordings.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- NVIDIA Canary 1B Flash had a word error rate of 9.85 percent on the GigaSpeech test set in the Open ASR Leaderboard table.
- 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.
- 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.
- NVIDIA Canary 1B Flash had a word error rate of 4.09 percent on Common Voice German.
- NVIDIA Canary 1B Flash had a word error rate of 6.99 percent on Common Voice English.
- NVIDIA Canary 1B Flash had a word error rate of 3.62 percent on Common Voice Spanish.
- NVIDIA Canary 1B Flash had a word error rate of 6.15 percent on Common Voice French.
- NVIDIA Canary 1B Flash had a word error rate of 4.36 percent on Multilingual LibriSpeech German.
- NVIDIA Canary 1B Flash had a word error rate of 2.69 percent on Multilingual LibriSpeech Spanish.
- NVIDIA Canary 1B Flash had a word error rate of 4.47 percent on Multilingual LibriSpeech French.
- 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.
- 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.
- NVIDIA Canary 1B Flash had a word error rate of 5.63 percent on the VoxPopuli test set in the Open ASR Leaderboard table.
- NVIDIA Canary 1B had a word error rate of 4.61 percent on Common Voice German.
- NVIDIA Canary 1B had a word error rate of 7.97 percent on Common Voice English.
- NVIDIA Canary 1B had a word error rate of 3.99 percent on Common Voice Spanish.
- NVIDIA Canary 1B had a word error rate of 6.53 percent on Common Voice French.
- NVIDIA Canary 1B had a word error rate of 4.19 percent on Multilingual LibriSpeech German.
- NVIDIA Canary 1B had a word error rate of 3.06 percent on Multilingual LibriSpeech English.
- NVIDIA Canary 1B had a word error rate of 3.15 percent on Multilingual LibriSpeech Spanish.
- NVIDIA Canary 1B had a word error rate of 4.12 percent on Multilingual LibriSpeech French.
- 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.
- NVIDIA Canary 1B v2 had an average word error rate of 8.85 percent on the CoVoST 2 test across its CoVoST-13 languages.
- NVIDIA Canary 1B v2 had a word error rate of 13.78 percent on the Earnings-22 long-form long-form recordings.
- 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.
- NVIDIA Canary 1B v2 had an average word error rate of 8.40 percent on the FLEURS test across its Fleurs-25 languages.
- NVIDIA Canary 1B v2 had a word error rate of 10.82 percent on the GigaSpeech test set in the Open ASR Leaderboard table.
- 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.
- 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.
- NVIDIA Canary 1B v2 had an average word error rate of 7.27 percent on the Multilingual LibriSpeech test across its MLS-6 languages.
- NVIDIA Canary 1B v2 had a word error rate of 5.08 percent on LibriSpeech Clean with MUSAN noise at 0 dB SNR.
- NVIDIA Canary 1B v2 had a word error rate of 2.29 percent on LibriSpeech Clean with MUSAN noise at 10 dB SNR.
- NVIDIA Canary 1B v2 had a word error rate of 2.80 percent on LibriSpeech Clean with MUSAN noise at 5 dB SNR.
- NVIDIA Canary 1B v2 had a word error rate of 19.38 percent on LibriSpeech Clean with MUSAN noise at -5 dB SNR.
- 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.
- 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.
- NVIDIA Canary 1B v2 had a word error rate of 9.87 percent on the This American Life long-form recordings.
- NVIDIA Canary 1B v2 had a word error rate of 6.25 percent on the VoxPopuli test set in the Open ASR Leaderboard table.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 4.84 percent on the CoVoST 2 German test.
- NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 6.80 percent on the CoVoST 2 English test.
- NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 3.41 percent on the CoVoST 2 Spanish test.
- NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 22.04 percent on the CoVoST 2 Estonian test.
- NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 6.05 percent on the CoVoST 2 French test.
- NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 3.69 percent on the CoVoST 2 Italian test.
- NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 38.36 percent on the CoVoST 2 Latvian test.
- NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 6.50 percent on the CoVoST 2 Dutch test.
- NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 3.96 percent on the CoVoST 2 Portuguese test.
- NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 31.80 percent on the CoVoST 2 Slovenian test.
- NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 20.16 percent on the CoVoST 2 Swedish test.
- NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 5.10 percent on the CoVoST 2 Ukrainian test.
- 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.
- NVIDIA Parakeet TDT 0.6B v3 had an average word error rate of 11.97 percent on the FLEURS test across its European languages.
- NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 12.64 percent on the FLEURS Bulgarian test.
- NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 11.01 percent on the FLEURS Czech test.
- NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 18.41 percent on the FLEURS Danish test.
- NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 5.04 percent on the FLEURS German test.
- NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 20.70 percent on the FLEURS Greek test.
- NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 4.85 percent on the FLEURS English test.
- NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 3.45 percent on the FLEURS Spanish test.
- NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 17.73 percent on the FLEURS Estonian test.
- NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 13.21 percent on the FLEURS Finnish test.
- NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 5.15 percent on the FLEURS French test.
- NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 12.46 percent on the FLEURS Croatian test.
- NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 15.72 percent on the FLEURS Hungarian test.
- NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 3.00 percent on the FLEURS Italian test.
- NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 20.35 percent on the FLEURS Lithuanian test.
- NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 22.84 percent on the FLEURS Latvian test.
- NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 20.46 percent on the FLEURS Maltese test.
- NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 7.48 percent on the FLEURS Dutch test.
- NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 7.31 percent on the FLEURS Polish test.
- NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 4.76 percent on the FLEURS Portuguese test.
- NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 12.44 percent on the FLEURS Romanian test.
- NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 5.51 percent on the FLEURS Russian test.
- NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 8.82 percent on the FLEURS Slovak test.
- NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 24.03 percent on the FLEURS Slovenian test.
- NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 15.08 percent on the FLEURS Swedish test.
- NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 6.79 percent on the FLEURS Ukrainian test.
- 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.
- 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.
- 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.
- 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.
- NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 4.39 percent on the Multilingual LibriSpeech Spanish test.
- NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 4.97 percent on the Multilingual LibriSpeech French test.
- NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 10.08 percent on the Multilingual LibriSpeech Italian test.
- NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 12.78 percent on the Multilingual LibriSpeech Dutch test.
- NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 7.28 percent on the Multilingual LibriSpeech Polish test.
- NVIDIA Parakeet TDT 0.6B v3 had a word error rate of 7.50 percent on the Multilingual LibriSpeech Portuguese test.
- 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.
- 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.
- 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.
- 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.
- NVIDIA Parakeet TDT 0.6B v3 had an average word error rate of 6.34 percent across the Open ASR Leaderboard's English datasets.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- In the same IBM study, the best human transcriber reached a word error rate of 6.8 percent on the CallHome telephone test.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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.
- In the Voxtral paper's Table 3, ElevenLabs Scribe had a word error rate of 8.35 percent on CHiME-4.
- In the Voxtral paper's Table 5, ElevenLabs Scribe had a word error rate of 47.03 percent on Common Voice Arabic.
- In the Voxtral paper's Table 5, ElevenLabs Scribe had a word error rate of 3.52 percent on Common Voice German.
- In the Voxtral paper's Table 5, ElevenLabs Scribe had a word error rate of 6.59 percent on Common Voice English.
- In the Voxtral paper's Table 5, ElevenLabs Scribe had a word error rate of 3.27 percent on Common Voice Spanish.
- In the Voxtral paper's Table 5, ElevenLabs Scribe had a word error rate of 17.29 percent on Common Voice Hindi.
- In the Voxtral paper's Table 5, ElevenLabs Scribe had a word error rate of 2.99 percent on Common Voice Italian.
- In the Voxtral paper's Table 5, ElevenLabs Scribe had a word error rate of 2.38 percent on Common Voice Dutch.
- In the Voxtral paper's Table 5, ElevenLabs Scribe had a word error rate of 5.46 percent on Common Voice Portuguese.
- In the Voxtral paper's Table 3, ElevenLabs Scribe had a word error rate of 7.39 percent on Earnings-21 (10-minute segments).
- In the Voxtral paper's Table 3, ElevenLabs Scribe had a word error rate of 9.16 percent on Earnings-22 (10-minute segments).
- In the Voxtral paper's Table 4, ElevenLabs Scribe had a word error rate of 11.58 percent on FLEURS Arabic.
- In the Voxtral paper's Table 4, ElevenLabs Scribe had a word error rate of 4.78 percent on FLEURS German.
- In the Voxtral paper's Table 4, ElevenLabs Scribe had a word error rate of 3.29 percent on FLEURS English.
- In the Voxtral paper's Table 4, ElevenLabs Scribe had a word error rate of 3.13 percent on FLEURS Spanish.
- In the Voxtral paper's Table 4, ElevenLabs Scribe had a word error rate of 5.07 percent on FLEURS French.
- In the Voxtral paper's Table 4, ElevenLabs Scribe had a word error rate of 5.67 percent on FLEURS Hindi.
- In the Voxtral paper's Table 4, ElevenLabs Scribe had a word error rate of 1.48 percent on FLEURS Italian.
- In the Voxtral paper's Table 4, ElevenLabs Scribe had a word error rate of 4.63 percent on FLEURS Dutch.
- In the Voxtral paper's Table 4, ElevenLabs Scribe had a word error rate of 4.50 percent on FLEURS Portuguese.
- In the Voxtral paper's Table 3, ElevenLabs Scribe had a word error rate of 10.52 percent on GigaSpeech.
- In the Voxtral paper's Table 3, ElevenLabs Scribe had a word error rate of 1.80 percent on LibriSpeech test-clean.
- In the Voxtral paper's Table 3, ElevenLabs Scribe had a word error rate of 3.44 percent on LibriSpeech test-other.
- In the Voxtral paper's Table 6, ElevenLabs Scribe had a word error rate of 9.81 percent on Multilingual LibriSpeech German.
- In the Voxtral paper's Table 6, ElevenLabs Scribe had a word error rate of 8.97 percent on Multilingual LibriSpeech Spanish.
- In the Voxtral paper's Table 6, ElevenLabs Scribe had a word error rate of 5.80 percent on Multilingual LibriSpeech French.
- In the Voxtral paper's Table 6, ElevenLabs Scribe had a word error rate of 12.38 percent on Multilingual LibriSpeech Italian.
- In the Voxtral paper's Table 6, ElevenLabs Scribe had a word error rate of 38.01 percent on Multilingual LibriSpeech Dutch.
- In the Voxtral paper's Table 6, ElevenLabs Scribe had a word error rate of 15.40 percent on Multilingual LibriSpeech Portuguese.
- In the Voxtral paper's Table 3, ElevenLabs Scribe had a word error rate of 3.16 percent on SPGI Speech.
- In the Voxtral paper's Table 3, ElevenLabs Scribe had a word error rate of 10.62 percent on Switchboard.
- In the Voxtral paper's Table 3, ElevenLabs Scribe had a word error rate of 6.95 percent on VoxPopuli.
- 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.
- In the Voxtral paper's Table 3, Gemini 2.5 Flash had a word error rate of 14.79 percent on CHiME-4.
- In the Voxtral paper's Table 5, Gemini 2.5 Flash had a word error rate of 53.62 percent on Common Voice Arabic.
- In the Voxtral paper's Table 5, Gemini 2.5 Flash had a word error rate of 7.25 percent on Common Voice German.
- In the Voxtral paper's Table 5, Gemini 2.5 Flash had a word error rate of 12.31 percent on Common Voice English.
- In the Voxtral paper's Table 5, Gemini 2.5 Flash had a word error rate of 5.88 percent on Common Voice Spanish.
- In the Voxtral paper's Table 5, Gemini 2.5 Flash had a word error rate of 11.86 percent on Common Voice French.
- In the Voxtral paper's Table 5, Gemini 2.5 Flash had a word error rate of 11.09 percent on Common Voice Hindi.
- In the Voxtral paper's Table 5, Gemini 2.5 Flash had a word error rate of 6.64 percent on Common Voice Italian.
- In the Voxtral paper's Table 5, Gemini 2.5 Flash had a word error rate of 5.73 percent on Common Voice Dutch.
- In the Voxtral paper's Table 5, Gemini 2.5 Flash had a word error rate of 9.42 percent on Common Voice Portuguese.
- 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).
- 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).
- 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.
- In the Voxtral paper's Table 4, Gemini 2.5 Flash had a word error rate of 4.74 percent on FLEURS German.
- In the Voxtral paper's Table 4, Gemini 2.5 Flash had a word error rate of 4.64 percent on FLEURS English.
- In the Voxtral paper's Table 4, Gemini 2.5 Flash had a word error rate of 3.17 percent on FLEURS Spanish.
- In the Voxtral paper's Table 4, Gemini 2.5 Flash had a word error rate of 6.17 percent on FLEURS French.
- In the Voxtral paper's Table 4, Gemini 2.5 Flash had a word error rate of 2.21 percent on FLEURS Italian.
- In the Voxtral paper's Table 4, Gemini 2.5 Flash had a word error rate of 6.20 percent on FLEURS Dutch.
- In the Voxtral paper's Table 4, Gemini 2.5 Flash had a word error rate of 4.23 percent on FLEURS Portuguese.
- In the Voxtral paper's Table 3, Gemini 2.5 Flash had a word error rate of 10.99 percent on GigaSpeech.
- In the same Voxtral table, Gemini 2.5 Flash had a word error rate of 6.76 percent in Hindi.
- In the Voxtral paper's Table 3, Gemini 2.5 Flash had a word error rate of 2.97 percent on LibriSpeech test-clean.
- In the Voxtral paper's Table 3, Gemini 2.5 Flash had a word error rate of 6.15 percent on LibriSpeech test-other.
- In the Voxtral paper's Table 6, Gemini 2.5 Flash had a word error rate of 6.52 percent on Multilingual LibriSpeech German.
- In the Voxtral paper's Table 6, Gemini 2.5 Flash had a word error rate of 4.39 percent on Multilingual LibriSpeech Spanish.
- In the Voxtral paper's Table 6, Gemini 2.5 Flash had a word error rate of 6.82 percent on Multilingual LibriSpeech French.
- In the Voxtral paper's Table 6, Gemini 2.5 Flash had a word error rate of 10.97 percent on Multilingual LibriSpeech Italian.
- In the Voxtral paper's Table 6, Gemini 2.5 Flash had a word error rate of 8.33 percent on Multilingual LibriSpeech Dutch.
- In the Voxtral paper's Table 6, Gemini 2.5 Flash had a word error rate of 7.14 percent on Multilingual LibriSpeech Portuguese.
- In the Voxtral paper's Table 3, Gemini 2.5 Flash had a word error rate of 4.00 percent on SPGI Speech.
- In the Voxtral paper's Table 3, Gemini 2.5 Flash had a word error rate of 9.57 percent on Switchboard.
- In the Voxtral paper's Table 3, Gemini 2.5 Flash had a word error rate of 7.84 percent on VoxPopuli.
- In the same Voxtral table, GPT-4o mini Transcribe had a word error rate of 14.02 percent in Arabic.
- In the Voxtral paper's Table 3, GPT-4o mini Transcribe had a word error rate of 11.35 percent on CHiME-4.
- In the Voxtral paper's Table 5, GPT-4o mini Transcribe had a word error rate of 51.68 percent on Common Voice Arabic.
- In the Voxtral paper's Table 5, GPT-4o mini Transcribe had a word error rate of 6.72 percent on Common Voice German.
- In the Voxtral paper's Table 5, GPT-4o mini Transcribe had a word error rate of 13.15 percent on Common Voice English.
- In the Voxtral paper's Table 5, GPT-4o mini Transcribe had a word error rate of 6.10 percent on Common Voice Spanish.
- In the Voxtral paper's Table 5, GPT-4o mini Transcribe had a word error rate of 10.75 percent on Common Voice French.
- In the Voxtral paper's Table 5, GPT-4o mini Transcribe had a word error rate of 39.06 percent on Common Voice Hindi.
- In the Voxtral paper's Table 5, GPT-4o mini Transcribe had a word error rate of 6.16 percent on Common Voice Italian.
- In the Voxtral paper's Table 5, GPT-4o mini Transcribe had a word error rate of 6.47 percent on Common Voice Dutch.
- In the Voxtral paper's Table 5, GPT-4o mini Transcribe had a word error rate of 9.95 percent on Common Voice Portuguese.
- 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).
- 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).
- In the Voxtral paper's Table 4, GPT-4o mini Transcribe had a word error rate of 3.76 percent on FLEURS German.
- In the Voxtral paper's Table 4, GPT-4o mini Transcribe had a word error rate of 3.19 percent on FLEURS English.
- In the Voxtral paper's Table 4, GPT-4o mini Transcribe had a word error rate of 2.58 percent on FLEURS Spanish.
- In the Voxtral paper's Table 4, GPT-4o mini Transcribe had a word error rate of 4.51 percent on FLEURS French.
- In the Voxtral paper's Table 4, GPT-4o mini Transcribe had a word error rate of 2.02 percent on FLEURS Italian.
- In the Voxtral paper's Table 4, GPT-4o mini Transcribe had a word error rate of 5.54 percent on FLEURS Dutch.
- In the Voxtral paper's Table 4, GPT-4o mini Transcribe had a word error rate of 3.54 percent on FLEURS Portuguese.
- In the Voxtral paper's Table 3, GPT-4o mini Transcribe had a word error rate of 14.80 percent on GigaSpeech.
- In the same Voxtral table, GPT-4o mini Transcribe had a word error rate of 12.36 percent in Hindi.
- In the Voxtral paper's Table 3, GPT-4o mini Transcribe had a word error rate of 1.92 percent on LibriSpeech test-clean.
- In the Voxtral paper's Table 3, GPT-4o mini Transcribe had a word error rate of 4.70 percent on LibriSpeech test-other.
- In the Voxtral paper's Table 6, GPT-4o mini Transcribe had a word error rate of 5.44 percent on Multilingual LibriSpeech German.
- In the Voxtral paper's Table 6, GPT-4o mini Transcribe had a word error rate of 4.28 percent on Multilingual LibriSpeech Spanish.
- In the Voxtral paper's Table 6, GPT-4o mini Transcribe had a word error rate of 4.77 percent on Multilingual LibriSpeech French.
- In the Voxtral paper's Table 6, GPT-4o mini Transcribe had a word error rate of 10.68 percent on Multilingual LibriSpeech Italian.
- In the Voxtral paper's Table 6, GPT-4o mini Transcribe had a word error rate of 8.62 percent on Multilingual LibriSpeech Dutch.
- In the Voxtral paper's Table 3, GPT-4o mini Transcribe had a word error rate of 17.31 percent on Switchboard.
- In the Voxtral paper's Table 3, GPT-4o mini Transcribe had a word error rate of 7.34 percent on VoxPopuli.
- In the Voxtral paper's Table 3, Voxtral Mini had a word error rate of 10.59 percent on CHiME-4.
- In the Voxtral paper's Table 5, Voxtral Mini had a word error rate of 63.98 percent on Common Voice Arabic.
- In the Voxtral paper's Table 5, Voxtral Mini had a word error rate of 6.07 percent on Common Voice German.
- In the Voxtral paper's Table 5, Voxtral Mini had a word error rate of 10.22 percent on Common Voice English.
- In the Voxtral paper's Table 5, Voxtral Mini had a word error rate of 4.98 percent on Common Voice Spanish.
- In the Voxtral paper's Table 5, Voxtral Mini had a word error rate of 8.92 percent on Common Voice French.
- In the Voxtral paper's Table 5, Voxtral Mini had a word error rate of 12.74 percent on Common Voice Hindi.
- In the Voxtral paper's Table 5, Voxtral Mini had a word error rate of 5.91 percent on Common Voice Italian.
- In the Voxtral paper's Table 5, Voxtral Mini had a word error rate of 6.03 percent on Common Voice Dutch.
- In the Voxtral paper's Table 5, Voxtral Mini had a word error rate of 7.76 percent on Common Voice Portuguese.
- In the Voxtral paper's Table 3, Voxtral Mini had a word error rate of 9.62 percent on Earnings-21 (10-minute segments).
- In the Voxtral paper's Table 3, Voxtral Mini had a word error rate of 11.84 percent on Earnings-22 (10-minute segments).
- In the Voxtral paper's Table 4, Voxtral Mini had a word error rate of 25.40 percent on FLEURS Arabic.
- In the Voxtral paper's Table 4, Voxtral Mini had a word error rate of 4.40 percent on FLEURS German.
- In the Voxtral paper's Table 4, Voxtral Mini had a word error rate of 3.77 percent on FLEURS English.
- In the Voxtral paper's Table 4, Voxtral Mini had a word error rate of 3.52 percent on FLEURS Spanish.
- In the Voxtral paper's Table 4, Voxtral Mini had a word error rate of 4.87 percent on FLEURS French.
- In the Voxtral paper's Table 4, Voxtral Mini had a word error rate of 9.26 percent on FLEURS Hindi.
- In the Voxtral paper's Table 4, Voxtral Mini had a word error rate of 2.51 percent on FLEURS Italian.
- In the Voxtral paper's Table 4, Voxtral Mini had a word error rate of 6.27 percent on FLEURS Dutch.
- In the Voxtral paper's Table 4, Voxtral Mini had a word error rate of 3.76 percent on FLEURS Portuguese.
- In the Voxtral paper's Table 3, Voxtral Mini had a word error rate of 10.68 percent on GigaSpeech.
- In the Voxtral paper's Table 3, Voxtral Mini had a word error rate of 1.86 percent on LibriSpeech test-clean.
- In the Voxtral paper's Table 3, Voxtral Mini had a word error rate of 4.04 percent on LibriSpeech test-other.
- In the Voxtral paper's Table 6, Voxtral Mini had a word error rate of 7.09 percent on Multilingual LibriSpeech German.
- In the Voxtral paper's Table 6, Voxtral Mini had a word error rate of 5.12 percent on Multilingual LibriSpeech Spanish.
- In the Voxtral paper's Table 6, Voxtral Mini had a word error rate of 5.28 percent on Multilingual LibriSpeech French.
- In the Voxtral paper's Table 6, Voxtral Mini had a word error rate of 11.30 percent on Multilingual LibriSpeech Italian.
- In the Voxtral paper's Table 6, Voxtral Mini had a word error rate of 10.09 percent on Multilingual LibriSpeech Dutch.
- In the Voxtral paper's Table 6, Voxtral Mini had a word error rate of 6.72 percent on Multilingual LibriSpeech Portuguese.
- 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.
- 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.
- 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.
- In the Voxtral paper's Table 3, Voxtral Mini had a word error rate of 2.19 percent on SPGI Speech.
- In the Voxtral paper's Table 3, Voxtral Mini had a word error rate of 11.32 percent on Switchboard.
- In the Voxtral paper's Table 3, Voxtral Mini Transcribe had a word error rate of 10.03 percent on CHiME-4.
- In the Voxtral paper's Table 5, Voxtral Mini Transcribe had a word error rate of 62.01 percent on Common Voice Arabic.
- In the Voxtral paper's Table 5, Voxtral Mini Transcribe had a word error rate of 4.85 percent on Common Voice German.
- In the Voxtral paper's Table 5, Voxtral Mini Transcribe had a word error rate of 8.25 percent on Common Voice English.
- In the Voxtral paper's Table 5, Voxtral Mini Transcribe had a word error rate of 3.96 percent on Common Voice Spanish.
- In the Voxtral paper's Table 5, Voxtral Mini Transcribe had a word error rate of 7.29 percent on Common Voice French.
- In the Voxtral paper's Table 5, Voxtral Mini Transcribe had a word error rate of 10.42 percent on Common Voice Hindi.
- In the Voxtral paper's Table 5, Voxtral Mini Transcribe had a word error rate of 4.37 percent on Common Voice Italian.
- In the Voxtral paper's Table 5, Voxtral Mini Transcribe had a word error rate of 4.71 percent on Common Voice Dutch.
- In the Voxtral paper's Table 5, Voxtral Mini Transcribe had a word error rate of 6.70 percent on Common Voice Portuguese.
- 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).
- 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).
- In the Voxtral paper's Table 4, Voxtral Mini Transcribe had a word error rate of 14.64 percent on FLEURS Arabic.
- In the Voxtral paper's Table 4, Voxtral Mini Transcribe had a word error rate of 3.54 percent on FLEURS German.
- In the Voxtral paper's Table 4, Voxtral Mini Transcribe had a word error rate of 3.61 percent on FLEURS English.
- In the Voxtral paper's Table 4, Voxtral Mini Transcribe had a word error rate of 2.75 percent on FLEURS Spanish.
- In the Voxtral paper's Table 4, Voxtral Mini Transcribe had a word error rate of 4.22 percent on FLEURS French.
- In the Voxtral paper's Table 4, Voxtral Mini Transcribe had a word error rate of 10.32 percent on FLEURS Hindi.
- In the Voxtral paper's Table 4, Voxtral Mini Transcribe had a word error rate of 2.31 percent on FLEURS Italian.
- In the Voxtral paper's Table 4, Voxtral Mini Transcribe had a word error rate of 4.89 percent on FLEURS Dutch.
- In the Voxtral paper's Table 4, Voxtral Mini Transcribe had a word error rate of 3.57 percent on FLEURS Portuguese.
- In the Voxtral paper's Table 3, Voxtral Mini Transcribe had a word error rate of 10.04 percent on GigaSpeech.
- In the Voxtral paper's Table 3, Voxtral Mini Transcribe had a word error rate of 1.57 percent on LibriSpeech test-clean.
- In the Voxtral paper's Table 3, Voxtral Mini Transcribe had a word error rate of 3.21 percent on LibriSpeech test-other.
- In the Voxtral paper's Table 6, Voxtral Mini Transcribe had a word error rate of 5.64 percent on Multilingual LibriSpeech German.
- In the Voxtral paper's Table 6, Voxtral Mini Transcribe had a word error rate of 3.87 percent on Multilingual LibriSpeech Spanish.
- In the Voxtral paper's Table 6, Voxtral Mini Transcribe had a word error rate of 4.14 percent on Multilingual LibriSpeech French.
- In the Voxtral paper's Table 6, Voxtral Mini Transcribe had a word error rate of 9.28 percent on Multilingual LibriSpeech Italian.
- In the Voxtral paper's Table 6, Voxtral Mini Transcribe had a word error rate of 9.63 percent on Multilingual LibriSpeech Dutch.
- In the Voxtral paper's Table 6, Voxtral Mini Transcribe had a word error rate of 5.17 percent on Multilingual LibriSpeech Portuguese.
- In the Voxtral paper's Table 3, Voxtral Mini Transcribe had a word error rate of 2.04 percent on SPGI Speech.
- In the Voxtral paper's Table 3, Voxtral Mini Transcribe had a word error rate of 11.35 percent on Switchboard.
- In the Voxtral paper's Table 3, Voxtral Mini had a word error rate of 6.85 percent on VoxPopuli.
- In the Voxtral paper's Table 3, Voxtral Small had a word error rate of 9.64 percent on CHiME-4.
- In the Voxtral paper's Table 5, Voxtral Small had a word error rate of 61.97 percent on Common Voice Arabic.
- In the Voxtral paper's Table 5, Voxtral Small had a word error rate of 3.74 percent on Common Voice German.
- In the Voxtral paper's Table 5, Voxtral Small had a word error rate of 8.58 percent on Common Voice English.
- In the Voxtral paper's Table 5, Voxtral Small had a word error rate of 3.31 percent on Common Voice Spanish.
- In the Voxtral paper's Table 5, Voxtral Small had a word error rate of 6.18 percent on Common Voice French.
- In the Voxtral paper's Table 5, Voxtral Small had a word error rate of 9.01 percent on Common Voice Hindi.
- In the Voxtral paper's Table 5, Voxtral Small had a word error rate of 3.98 percent on Common Voice Dutch.
- In the Voxtral paper's Table 5, Voxtral Small had a word error rate of 6.43 percent on Common Voice Portuguese.
- In the Voxtral paper's Table 3, Voxtral Small had a word error rate of 9.55 percent on Earnings-21 (10-minute segments).
- In the Voxtral paper's Table 3, Voxtral Small had a word error rate of 12.48 percent on Earnings-22 (10-minute segments).
- In the Voxtral paper's Table 4, Voxtral Small had a word error rate of 13.44 percent on FLEURS Arabic.
- In the Voxtral paper's Table 4, Voxtral Small had a word error rate of 3.38 percent on FLEURS German.
- In the Voxtral paper's Table 4, Voxtral Small had a word error rate of 3.35 percent on FLEURS English.
- In the Voxtral paper's Table 4, Voxtral Small had a word error rate of 2.72 percent on FLEURS Spanish.
- In the Voxtral paper's Table 4, Voxtral Small had a word error rate of 4.03 percent on FLEURS French.
- In the Voxtral paper's Table 4, Voxtral Small had a word error rate of 7.69 percent on FLEURS Hindi.
- In the Voxtral paper's Table 4, Voxtral Small had a word error rate of 2.62 percent on FLEURS Italian.
- In the Voxtral paper's Table 4, Voxtral Small had a word error rate of 4.94 percent on FLEURS Dutch.
- In the Voxtral paper's Table 4, Voxtral Small had a word error rate of 3.79 percent on FLEURS Portuguese.
- In the Voxtral paper's Table 3, Voxtral Small had a word error rate of 10.27 percent on GigaSpeech.
- In the Voxtral paper's Table 3, Voxtral Small had a word error rate of 1.53 percent on LibriSpeech test-clean.
- In the Voxtral paper's Table 3, Voxtral Small had a word error rate of 3.14 percent on LibriSpeech test-other.
- In the Voxtral paper's Table 6, Voxtral Small had a word error rate of 5.57 percent on Multilingual LibriSpeech German.
- In the Voxtral paper's Table 6, Voxtral Small had a word error rate of 3.62 percent on Multilingual LibriSpeech Spanish.
- In the Voxtral paper's Table 6, Voxtral Small had a word error rate of 3.73 percent on Multilingual LibriSpeech French.
- In the Voxtral paper's Table 6, Voxtral Small had a word error rate of 8.44 percent on Multilingual LibriSpeech Italian.
- In the Voxtral paper's Table 6, Voxtral Small had a word error rate of 9.43 percent on Multilingual LibriSpeech Dutch.
- In the Voxtral paper's Table 6, Voxtral Small had a word error rate of 5.85 percent on Multilingual LibriSpeech Portuguese.
- 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.
- 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.
- 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.
- In the Voxtral paper's Table 3, Voxtral Small had a word error rate of 1.89 percent on SPGI Speech.
- In the Voxtral paper's Table 3, Voxtral Small had a word error rate of 11.09 percent on Switchboard.
- In the Voxtral paper's Table 3, Voxtral Small had a word error rate of 6.62 percent on VoxPopuli.
- In the Voxtral paper's FLEURS table, Whisper large-v3 had a word error rate of 15.44 percent in Arabic.
- In the Voxtral paper's Table 3, Whisper large-v3 had a word error rate of 10.88 percent on CHiME-4.
- In the Voxtral paper's Table 5, Whisper large-v3 had a word error rate of 50.58 percent on Common Voice Arabic.
- In the Voxtral paper's Table 5, Whisper large-v3 had a word error rate of 6.25 percent on Common Voice German.
- In the Voxtral paper's Table 5, Whisper large-v3 had a word error rate of 22.91 percent on Common Voice English.
- In the Voxtral paper's Table 5, Whisper large-v3 had a word error rate of 5.66 percent on Common Voice Spanish.
- In the Voxtral paper's Table 5, Whisper large-v3 had a word error rate of 11.33 percent on Common Voice French.
- On the Mozilla Common Voice Hindi test in the Voxtral paper, Whisper large-v3 had a word error rate of 46.75 percent.
- In the Voxtral paper's Table 5, Whisper large-v3 had a word error rate of 6.81 percent on Common Voice Italian.
- In the Voxtral paper's Table 5, Whisper large-v3 had a word error rate of 5.83 percent on Common Voice Dutch.
- In the Voxtral paper's Table 5, Whisper large-v3 had a word error rate of 7.17 percent on Common Voice Portuguese.
- 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).
- 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).
- In the Voxtral paper's Table 4, Whisper large-v3 had a word error rate of 5.46 percent on FLEURS German.
- In the Voxtral paper's Table 4, Whisper large-v3 had a word error rate of 4.00 percent on FLEURS English.
- In the Voxtral paper's Table 4, Whisper large-v3 had a word error rate of 2.81 percent on FLEURS Spanish.
- In the Voxtral paper's Table 4, Whisper large-v3 had a word error rate of 5.55 percent on FLEURS French.
- In the Voxtral paper's Table 4, Whisper large-v3 had a word error rate of 2.71 percent on FLEURS Italian.
- In the Voxtral paper's Table 4, Whisper large-v3 had a word error rate of 5.87 percent on FLEURS Dutch.
- In the Voxtral paper's Table 4, Whisper large-v3 had a word error rate of 3.90 percent on FLEURS Portuguese.
- In the Voxtral paper's Table 3, Whisper large-v3 had a word error rate of 11.60 percent on GigaSpeech.
- In the Voxtral paper's FLEURS table, Whisper large-v3 had a word error rate of 28.87 percent in Hindi.
- In the Voxtral paper's Table 3, Whisper large-v3 had a word error rate of 1.84 percent on LibriSpeech test-clean.
- In the Voxtral paper's Table 3, Whisper large-v3 had a word error rate of 3.66 percent on LibriSpeech test-other.
- In the Voxtral paper's Table 6, Whisper large-v3 had a word error rate of 5.72 percent on Multilingual LibriSpeech German.
- In the Voxtral paper's Table 6, Whisper large-v3 had a word error rate of 3.89 percent on Multilingual LibriSpeech Spanish.
- In the Voxtral paper's Table 6, Whisper large-v3 had a word error rate of 5.09 percent on Multilingual LibriSpeech French.
- In the Voxtral paper's Table 6, Whisper large-v3 had a word error rate of 9.78 percent on Multilingual LibriSpeech Italian.
- In the Voxtral paper's Table 6, Whisper large-v3 had a word error rate of 9.19 percent on Multilingual LibriSpeech Dutch.
- In the Voxtral paper's Table 6, Whisper large-v3 had a word error rate of 7.03 percent on Multilingual LibriSpeech Portuguese.
- In the Voxtral paper's Table 3, Whisper large-v3 had a word error rate of 3.15 percent on SPGI Speech.
- In the Voxtral paper's Table 3, Whisper large-v3 had a word error rate of 13.14 percent on Switchboard.
- In the Voxtral paper's Table 3, Whisper large-v3 had a word error rate of 9.58 percent on VoxPopuli.
- 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.
- 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.
- On the CallHome test set of conversations between friends and family, Whisper Large V2 had a word error rate of 17.6 percent.
- On the AMI meeting corpus recorded with individual headset microphones (IHM), Whisper Large V2 had a word error rate of 16.9 percent.
- On the AMI meeting corpus recorded with a single distant microphone (SDM1), Whisper Large V2 had a word error rate of 36.4 percent.
- On the CHiME6 test set, Whisper Large V2 had a word error rate of 25.5 percent.
- On the Common Voice 9 test set, the Whisper large-v2 model had a word error rate of 53.8 percent for Arabic.
- On the Common Voice 9 test set, the Whisper large-v2 model had a word error rate of 39.4 percent for Persian.
- On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 36.7 percent for Afrikaans.
- 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.
- On the Fleurs benchmark, the Whisper large-v2 model had a word error rate of 16.0 percent for Arabic.
- On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 44.6 percent for Armenian.
- On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 106.2 percent for Assamese.
- On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 23.4 percent for Azerbaijani.
- On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 45.4 percent for Belarusian.
- On the Fleurs benchmark, the Whisper large-v2 model had a word error rate of 104.1 percent for Bengali.
- On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 15.7 percent for Bosnian.
- On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 14.6 percent for Bulgarian.
- On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 7.3 percent for Catalan.
- On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 13.4 percent for Croatian.
- On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 13.3 percent for Czech.
- On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 13.8 percent for Danish.
- On the FLEURS read-speech benchmark, Whisper large-v2 had a word error rate of 6.7 percent in Dutch.
- On the Fleurs benchmark, the Whisper large-v2 model had a word error rate of 4.2 percent for English.
- On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 21.9 percent for Estonian.
- On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 9.7 percent for Finnish.
- On the Fleurs benchmark, the Whisper large-v2 model had a word error rate of 8.3 percent for French.
- On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 15.4 percent for Galician.
- On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 105.0 percent for Georgian.
- On the Fleurs benchmark, the Whisper large-v2 model had a word error rate of 4.5 percent for German.
- On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 12.5 percent for Greek.
- On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 102.7 percent for Gujarati.
- On the FLEURS benchmark, Whisper large-v2 had a word error rate of 88.9 percent for Hausa.
- On the Fleurs benchmark, the Whisper large-v2 model had a word error rate of 27.1 percent for Hebrew.
- On the Fleurs benchmark, the Whisper large-v2 model had a word error rate of 21.5 percent for Hindi.
- On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 17.0 percent for Hungarian.
- On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 38.2 percent for Icelandic.
- On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 7.1 percent for Indonesian.
- On the FLEURS read-speech benchmark, Whisper large-v2 had a word error rate of 4.0 percent in Italian.
- On the Fleurs benchmark, the Whisper large-v2 model had a word error rate of 37.0 percent for Kannada.
- On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 37.7 percent for Kazakh.
- On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 99.7 percent for Khmer.
- On the Fleurs benchmark, the Whisper large-v2 model had a word error rate of 14.3 percent for Korean.
- On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 23.1 percent for Latvian.
- On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 75.6 percent for Lingala.
- On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 28.1 percent for Lithuanian.
- On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 88.0 percent for Luxembourgish.
- On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 16.5 percent for Macedonian.
- On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 8.7 percent for Malay.
- On the Fleurs benchmark, the Whisper large-v2 model had a word error rate of 100.7 percent for Malayalam.
- On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 76.6 percent for Maltese.
- On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 38.5 percent for Maori.
- On the Fleurs benchmark, the Whisper large-v2 model had a word error rate of 38.3 percent for Marathi.
- On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 110.5 percent for Mongolian.
- On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 47.1 percent for Nepali.
- On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 9.5 percent for Norwegian.
- On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 75.3 percent for Occitan.
- On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 93.7 percent for Pashto.
- On the Fleurs benchmark, the Whisper large-v2 model had a word error rate of 32.9 percent for Persian.
- On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 5.4 percent for Polish.
- On the FLEURS read-speech benchmark, Whisper large-v2 had a word error rate of 4.3 percent in Portuguese.
- On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 102.4 percent for Punjabi.
- On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 14.4 percent for Romanian.
- On the Fleurs benchmark, the Whisper large-v2 model had a word error rate of 5.6 percent for Russian.
- On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 33.9 percent for Serbian.
- On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 121.0 percent for Shona.
- On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 156.5 percent for Sindhi.
- On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 11.7 percent for Slovak.
- On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 102.9 percent for Somali.
- On the Fleurs benchmark, the Whisper large-v2 model had a word error rate of 3.0 percent for Spanish.
- On the Fleurs benchmark, the Whisper large-v2 model had a word error rate of 39.3 percent for Swahili.
- On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 8.5 percent for Swedish.
- On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 85.8 percent for Tajik.
- On the Fleurs benchmark, the Whisper large-v2 model had a word error rate of 17.5 percent for Tamil.
- On the Fleurs benchmark, the Whisper large-v2 model had a word error rate of 99.0 percent for Telugu.
- On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 8.4 percent for Turkish.
- On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 8.6 percent for Ukrainian.
- On the Fleurs benchmark, the Whisper large-v2 model had a word error rate of 22.6 percent for Urdu.
- On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 90.2 percent for Uzbek.
- On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 10.3 percent for Vietnamese.
- On the FLEURS benchmark, the Whisper large-v2 model had a word error rate of 33.0 percent for Welsh.
- On the Fleurs benchmark, the Whisper large-v2 model had a word error rate of 94.8 percent for Yoruba.
- On the LibriSpeech Clean test set, Whisper Large V2 had a word error rate of 2.7 percent.
- 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.
- On the Switchboard test set, Whisper Large V2 had a word error rate of 13.8 percent.
- 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.
- 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.
- 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.