Fact · Speech recognition
1.60percent word error rate
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 found word for word on the live source, last checked October 19, 2026
In the source's own words
“| **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 |”
canary-qwen-2.5b model card (raw README), NVIDIA, Hugging Face (NVIDIA model card), 2025, Open ASR Leaderboard table, LibriSpeech test-clean column.
Open the original- What it measures
- The model card's Hugging Face Open ASR Leaderboard table gives the WER for each dataset in the leaderboard's English set.
- What it does not show
- A vendor-run score on the leaderboard's English datasets, with the leaderboard's text normalization. Compare only with systems scored the same way.
In the data
A row of Speech Recognition Accuracy: Word Error Rates, as Canary-Qwen 2.5B, LS test-clean English: 1.6 percent.
- System
- NVIDIA Canary Qwen 2.5B
- Kind
- Open model
- Test set
- LibriSpeech test-clean
- Language
- English
- Speech
- Clean read speech
- Year
- 2025