whatmodel.sh / stt / Vosk en-us 0.22
Vosk en-us 0.22
↗ share
+ contribute
Vosk en-us 0.22 is an open-weights speech-to-text model from Alpha Cephei (Vosk), released October 2021. Licence: apache-2.0.
Catalogue identity notes
Exact Vosk (Kaldi-based) English model 'vosk-model-en-us-0.22' from alphacephei.com/vosk/models (listed 1.8G, Apache 2.0; page's own WER 5.69 (librispeech test-clean) 6.05 (tedlium) 29.78 (callcenter)). Kaldi TDNN-style offline/streaming recogniser, the landmark pre-neural-end-to-end open ASR stack. released: not stated on the models page (left unset).
Model information sources
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 Vosk en-us 0.22, out of 176 with a score, then more from Alpha Cephei (Vosk).
| # | model | score | where | ||
|---|---|---|---|---|---|
| 144 |  STT En Conformer-Transducer Small | 43±3 | local | ||
| 145 |  ESPnet Conformer 2020 + RNN LM | 43±4 | local | ||
| 146 | Vosk en-us 0.22 | 42±6 | local | ||
| 147 |  GigaAM Multilingual (CTC 220M) | 42±7 | local | ||
| 148 | ESPnet Transformer 18-layer + RNN LM | 42±4 | local | ||
| more from Alpha Cephei (Vosk) | |||||
| 158 | Vosk small en-us 0.15 | 31±6 | 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 · 85% of its score
| test | result | vs the others in its run | counts |
|---|---|---|---|
| Picovoice STT benchmark · English streaming 4-set mean WER | 11.5% ✎ | █·█···▄▄▄··▄ beats 5/7 | 67% |
| LibriSpeech test-clean WER lab | 5.7% ✎ | ▁·▁·▁▃▁▁▃▇█▆ beats 5/100 | 14% |
| TED-LIUM test WER (short-form) lab | 6.0% ✎ | ▁·▁▇█▃·▁▃▃█▄ beats 16/27 | 3.3% |
$ where to run it
Other ways to run it
- Run it as published · needs ~16 GB · from github.com/alphacep (official)
uvx --from vosk vosk-transcriber -n vosk-model-en-us-0.22 -i audio.wav -o out.txt
Weights and documentation: alphacephei.com ↗
$ how the score was worked out · Picovoice speech-to-text-benc… 67%, Alpha Cephei Vosk models page… 18%
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 |
|---|---|---|
 Picovoice speech-to-text-benchmark · English batch-engine WER (README) | 1 | 67% |
| Alpha Cephei Vosk models page (vendor-reported WER for its own Vosk models) · lab-reported | 2 | 18% |
| prior · pull toward the middle | 15% | |
| weighted average | = 42.2 |
$ 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