Fact · Speech recognition
8.85percent word error rate
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 found word for word on the live source, last checked October 19, 2026
In the source's own words
“| **WER ↓** | Fleurs-25 Langs | CoVoST-13 Langs | MLS - 6 Langs | ... | **`Canary-1b-v2`** | 8.40% | 8.85% | 7.27% |”
canary-1b-v2 model card (raw README), NVIDIA, Hugging Face (NVIDIA model card), 2025, ASR table, CoVoST-13 column.
Open the original- What it measures
- The model card's ASR table gives the average WER over the languages each test set covers. The card excludes punctuation and capitalization errors.
- What it does not show
- NVIDIA's own average over a set of languages, not a per-language figure, so it hides large gaps between languages.
In the data
A row of Speech Recognition Accuracy: Word Error Rates, as Canary 1B v2, CoVoST 2: 8.85 percent.
- System
- NVIDIA Canary 1B v2
- Kind
- Open model
- Test set
- CoVoST 2
- Language
- European languages (average)
- Year
- 2025