Common Voice
Phi-4-multimodal 6.80% leads since Mar 2025 Fading Crowd-sourced multilingual read-speech corpus. Used here as an ASR test set scored by word error rate.
Mozilla releases numbered corpus versions such as 13.0 and 17.0, so reports should name the version. Since 2025-10 the datasets are only on Mozilla Data Collective, not Hugging Face.
In the authors’ words
The Common Voice corpus is a massively-multilingual collection of transcribed speech intended for speech technology research and development. Common Voice is designed for Automatic Speech Recognition purposes but can be useful in other domains (e.g. language identification). To achieve scale and sustainability, the Common Voice project employs crowdsourcing for both data collection and data validation. The most recent release includes 29 languages, and as of November 2019 there are a total of 38 languages collecting data. Over 50,000 individuals have participated so far, resulting in 2,500 hours of collected audio. To our knowledge this is the largest audio corpus in the public domain for speech recognition, both in terms of number of hours and number of languages. As an example use case for Common Voice, we present speech recognition experiments using Mozilla's DeepSpeech Speech-to-Text toolkit. By applying transfer learning from a source English model, we find an average Character Error Rate improvement of 5.99 +/- 5.48 for twelve target languages (German, French, Italian, Turkish, Catalan, Slovenian, Welsh, Irish, Breton, Tatar, Chuvash, and Kabyle). For most of these languages, these are the first ever published results on end-to-end Automatic Speech Recognition.
- Level
- Everyday Most adults, without training
- Scored
- Checked automatically A program checks the result against an answer key, unit tests or the expected final state.
- Answers
- Short answer A number, word or short string matched to a key.
Phi-4-multimodal leads with 6.80% among models released by Dec 2025
Best score per model as labs report them in their own cards. Labs pick their own settings, so these compare less cleanly than independent runs.
The newest model with a score here came out in Dec 2025, and nobody has reported a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6 Astra
- Claude Opus 5.5
- GPT-6.1 Terra
- GPT-6.1 Sol
- and 16 more
- 1 Phi-4-multimodal 6.80%
- 2
Nova 2 Omni Preview 5.70% - 3 Nova 2.0 Pro Preview 5.50%
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Phi-4-multimodal | 6.80% | Mar 2025 | Model card | |
| 2 | Nova 2 Omni Preview | 5.70% | Dec 2025 | Model card | |
| 3 | Nova 2.0 Pro Preview | 5.50% | Dec 2025 | Model card |
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Labs headline it less often now, last in Mar 2026
Launch posts from every lab that has listed it, one tick per post. Tall colored ticks include it, and short gray ones leave it out.
| Date | Launch post | Score | As the post writes it |
|---|---|---|---|
| 30 Mar 2026 |  Qwen3.5-Omni-Plus | 3.13 | Mean of CV15 zh 3.46 / yue 1.95 / zh-tw 2.27 / en 4.83 |
| 2 Dec 2025 | Nova 2 Omni Preview | 5.7 | Common Voice (v15) WER, five languages |
| 2 Dec 2025 | Nova 2.0 Pro Preview | 5.5 | Common Voice WER, EFIGS average |
| 31 Oct 2025 |  LongCat-Flash-Omni | 4.98 | CommonVoice 15 (zh | en) 4.98 | 13.59 |
| 22 Sep 2025 | Qwen3-Omni-30B-A3B | 5.18 | Mean of CV15-en 6.05 / CV15-zh 4.31 |
| 15 Jul 2025 |  Voxtral Small | Mozilla Common Voice, macro-average WER (chart) | |
| 27 Mar 2025 | Qwen2.5-Omni-7B | 6.9 | Mean of Common Voice 15 en 7.6 / zh 5.2 / yue 7.3 / fr 7.5 |
8 system cards report it, from 5 labs
Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.
| Date | Document | Score | As the document writes it | Section |
|---|---|---|---|---|
| 17 Apr 2026 | Qwen3.5-Omni-Plus technical report | CV15 ASR WER zh 3.46, yue 1.95, zh-tw 2.27, en 4.83, no overall (Table 5) | Capabilities | |
| 2 Dec 2025 | Nova 2 Omni Preview technical report | 5.7 | Common Voice (v15) WER, five languages, lower is better (Table 7) | Capabilities |
| 2 Dec 2025 | Nova 2.0 Pro Preview technical report | 5.5 | Common Voice WER, EFIGS average, lower is better (Table 3) | Capabilities |
| 31 Oct 2025 | LongCat-Flash-Omni technical report | CommonVoice 15 zh 4.98, en 13.59, error rate, lower is better, no overall score (Table 9) | Capabilities | |
| 22 Sep 2025 | Qwen3-Omni-30B-A3B technical report | CV15-en 6.05, CV15-zh 4.31, WER, lower is better, Instruct, no overall score (Table 6) | Capabilities | |
| 17 Jul 2025 | Voxtral Small technical report | Mozilla Common Voice 15.1 WER per language with no printed macro-average, lower is better: ar 61.97, nl 3.98, en 8.58, fr 6.18, de 3.74, hi 9.01, it 3.96, pt 6.43, es 3.31 (Table 5, Figure 3 chart) | Capabilities | |
| 26 Mar 2025 | Qwen2.5-Omni-7B technical report | Common Voice 15 en 7.6, zh 5.2, yue 7.3, fr 7.5, WER, lower is better, no overall score (Table 2) | Capabilities | |
| 3 Mar 2025 | Phi-4-multimodal technical report | 6.8 | CV15 (CommonVoice 15) ASR, WER average over 8 languages, lower is better (Tables 3 and 4) | Capabilities |
More in Audio and video
- Video-MMMU In use 300 expert-level lecture videos with questions testing perception, comprehension and adaptation of knowledge. Accuracy. Qwen 3.8 Max 88.7%
- LVBench In use Hours-long video understanding with multiple-choice questions on reasoning, grounding and summarization.
Gemini 4 Argon 91.7% - MMVU In use 3,000 expert-annotated questions on videos across 27 disciplines. Multiple-choice and open-ended accuracy. Qwen 3.8 Max 82.4%
- MLVU In use Multi-task long video understanding (minutes to hours) with multiple-choice and generation tasks. Accuracy. Qwen 3.8 Max 90.8%
- VoiceBench In use Spoken instructions with speaker, environment and content variation. Voice assistant responses scored by GPT judge or rules. Qwen3.5-Omni-Plus 93.1%
- TVBench In use Models answer multiple-choice questions on 10 temporal video tasks, such as action counting and ordering, that static frames cannot solve. Scored by accuracy. Qwen 3.8 Max 81.9%