whatmodel.sh / stt / data2vec Audio Large 960h
data2vec Audio Large 960h
↗ share
+ contribute
data2vec Audio Large 960h is an open-weights speech-to-text model from Meta, released April 2022. Its published size is 0.31B parameters. Licence: apache-2.0.
Catalogue identity notes
Exact checkpoint facebook/data2vec-audio-large-960h (CTC fine-tuned on LibriSpeech 960h). Params: fp32 pytorch_model.bin 1,253,377,393 bytes / 4; released: HF repo creation month.
Score within the Open ASR English short-form WER (March 2026) comparison set only.
$ how it compares
The speech-to-text models ranked either side of data2vec Audio Large 960h, out of 176 with a score, then more from Meta.
| # | model | score | where | ||
|---|---|---|---|---|---|
| 169 | wav2vec 2.0 Conformer Rel-Pos Large 960h | 14±4 | local | ||
| 170 | wav2vec 2.0 Conformer RoPE Large 960h | 13±4 | local | ||
| 171 | data2vec Audio Large 960h | 13±4 | local | ||
| 172 | wav2vec 2.0 Large 960h | 7±3 | local | ||
| 173 | data2vec Audio Base 960h | 4±4 | local | ||
| more from Meta | |||||
| 1 | Muse Voice Transcribe 1.0 | 71±3 | api $0.0030/audio min | ||
| 58 | Omnilingual ASR LLM 7B | 58±2 | local | ||
| 59 | Omnilingual ASR LLM 7B v2 | 58±3 | local | ||
see all speech-to-text models · find the ones that fit your machine
$ what it's good at
other tests3 not in a task · 95% of its score
| test | result | vs the others in its run | counts |
|---|---|---|---|
| Open ASR English short-form WER (March 2026) | 23.2% ✎ | ▁▁▁▃··▁▁▂██▆ beats 6/86 | 43% |
| LibriSpeech test-other WER lab | 4.1% ✎ | ▂·▂·▂······█ beats 3/6 | 38% |
| LibriSpeech test-clean WER lab | 1.9% ✎ | ▂··▂▂·····▂█ beats 6/8 | 13% |
$ where to run it
Other ways to run it
- Run it as published · needs ~1.99 GB · from huggingface.co/facebook
uv run --with transformers --with torch python -c "from transformers import pipeline; print(pipeline('automatic-speech-recognition', model='facebook/data2vec-audio-large-960h', chunk_length_s=30)('audio.wav')['text'])"
Weights and documentation: huggingface.co/facebook ↗
$ how the score was worked out · Meta (facebook org) Hugging F… 52%, Open ASR Leaderboard 43%
Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works
| source | results | share of the score |
|---|---|---|
 Meta (facebook org) Hugging Face ASR model cards: published LibriSpeech WER for its own checkpoints · lab-reported | 2 | 52% |
| Open ASR Leaderboard | 1 | 43% |
| prior · pull toward the middle | 5% | |
| weighted average | = 13.1 |
$ improve this page
Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.
data updated · sources and licences · json