speech to text benchmark
CoVoST2
CoVoST 2 is a large-scale multilingual speech translation corpus derived from Common Voice, covering translations from 21 languages into English and from English into 15 languages. The dataset contains 2,880 hours of speech with 78K speakers for speech translation research.
Updated Aug 7, 2026
Published models4
Registry coverage4
MetricScore
EvidenceB
CoVoST2 leaderboard
Sorted by the source-provided rank. Higher score is better according to the registry.
| 1 | AM Nova 2 OmniAmazon | 40.7% | 100.0% | 4 | C | |
| 2 | GO | 39.2% | 66.7% | 4 | C | |
| 3 | GO | 38.5% | 33.3% | 4 | C | |
| 4 | GO | 38.4% | 0.0% | 4 | C |
Score distribution
Top published rows on the benchmark's original scale.
CoVoST2 highlights
The top published results on this benchmark's own scale.
What is CoVoST2?
Definition and scoring fields from the benchmark registry.
CoVoST 2 is a large-scale multilingual speech translation corpus derived from Common Voice, covering translations from 21 languages into English and from English into 15 languages. The dataset contains 2,880 hours of speech with 78K speakers for speech translation research.
Scores are shown in ratio. The current registry marks this benchmark as not independently verified with evidence level B.
- Family
- CoVoST2
- Modality
- audio
- Primary category
- speech to text
- Score direction
- higher
- LLMBoard eligible
- No
- Evaluation key
- covost2|llm-stats-current
Source-native results are preserved. Eligibility for the overall LLMBoard score is a separate policy decision.
FAQ
Common questions about CoVoST2.
Which model scores highest on CoVoST2?
Nova 2 Omni is currently ranked first with 40.7%.
What does CoVoST2 measure?
CoVoST 2 is a large-scale multilingual speech translation corpus derived from Common Voice, covering translations from 21 languages into English and from English into 15 languages. The dataset contains 2,880 hours of speech with 78K speakers for speech translation research.
Is a higher score better?
Yes. Higher values rank better for this benchmark.
How many models are compared?
4 unique published model results are currently shown.
Does this benchmark affect the overall score?
This benchmark is preserved as source-native evidence but is not eligible for the current overall score.