whatmodel.sh / stt / OWSM-CTC v3.2 ft 1B

OWSM-CTC v3.2 ft 1B

+ contribute
Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

OWSM-CTC v3.2 ft 1B is an open-weights speech-to-text model from ESPnet (CMU WAVLab), released September 2024. Its published size is 1.01B parameters. Licence: cc-by-4.0.

Catalogue identity notes

Exact checkpoint espnet/owsm_ctc_v3.2_ft_1B. Params derived from the fp32 weight file (4020755920 bytes / 4 = 1.005B; card gives only '1B'). released: HF repo creation month. Command follows the card's Speech2Text.decode_long example (espnet needs Python 3.12-3.13).

49.1±4.1#106 of 162 speech-to-text modelsfrom 3 benchmark results across 2 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 OWSM-CTC v3.2 ft 1B, out of 176 with a score, then more from ESPnet (CMU WAVLab).

#modelscorewhereRAMreleased
116Omnilingual ASR CTC 1B v2 Meta50±3local~2.9GB2025-12
117Whisper large (v1) OpenAI49±2local~4.2GB2022-09
118OWSM-CTC v3.2 ft 1B ESPnet (CMU WAVLab)49±4local~5.4GB2024-09
119Nemotron Speech Streaming EN 0.6B NVIDIA49±4local~0.7GB2026-03
120Ultravox v0.6 (Llama 3.1 8B) Fixie.ai (Ultravox)49±5comparison only2025-06
more from ESPnet (CMU WAVLab)
103OWSM-CTC v4 1B ESPnet (CMU WAVLab)52±4local~5.4GB2025-01
114OWSM-CTC v3.1 1B ESPnet (CMU WAVLab)50±4local~5.4GB2024-02

see all speech-to-text models · find the ones that fit your machine

$ what it's good at

other tests3 not in a task · 93% of its score
testresultvs the others in its rundifficultycountswhy
Open ASR English short-form WER (March 2026)8.3% ✎▁▁▁▃··▁▁▂██▆ beats 31/86avg 8.2%▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 59%blunt
LibriSpeech test-other WER lab5.3% ✎▁▁▁▂▅▃▅▄█▂▂▁ beats 18/54avg 5.0%▇▇▇▇▇▇▇▇▇▇ 26%small run
LibriSpeech test-clean WER lab2.6% ✎▁·▁·▁▁▁▁▃▇█▆ beats 27/100avg 2.3%▇▇▇ 8.0%blunt · small run

$ where to run it

Other ways to run it

  1. Run it as published · needs ~5.35 GB, NVIDIA only · from huggingface.co/espnet (official)
    uv run --python 3.12 --with espnet --with espnet_model_zoo --with librosa python -c "from espnet2.bin.s2t_inference import Speech2Text; s=Speech2Text.from_pretrained('espnet/owsm_ctc_v3.2_ft_1B', device='cuda', lang_sym='<eng>', task_sym='<asr>'); print(' '.join(t for _, _, t in s.decode_long('audio.wav', batch_size=16)))"

Official weights and documentation: huggingface.co/espnet ↗

$ how the score was worked out · Open ASR Leaderboard 59%, OWSM v4 technical report (ESP… 34%

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

sourceresultsshare of the scorelicence
Open ASR Leaderboard1▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 59%Apache-2.0
OWSM v4 technical report (ESPnet / CMU WAVLab, arXiv:2506.00338v1): own OWSM-CTC scores · lab-reported2▇▇▇▇▇▇▇▇▇▇ 34%Factual vendor-reported score (not licensed content)
prior · pull toward the middle▇▇ 7%
weighted average= 49.1

$ improve this page

Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

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