whatmodel.sh / stt / ABR asr-19m-v2-en

ABR asr-19m-v2-en

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

ABR asr-19m-v2-en is an open-weights speech-to-text model from Applied Brain Research, released November 2025. Its published size is 0.019B parameters. Licence: applied-brain-research-open-license.

Catalogue identity notes

State-space ASR model; CSV entry 'abr-ai/asr-19m-v2-en-32b'. The repo was renamed (now abr-ai/niagara-19m-batch.en) and replaced with a 'New model version' on 2026-04-07, after this evaluation. The evaluated checkpoint is the one at revision dab6545 (2026-03-25), whose card model-index repeats these leaderboard WERs (AMI 18.76, Earnings22 13.53). No local variant: current downloads give the newer checkpoint. Params: card (19M).

40.8±5.1#136 of 162 speech-to-text modelsfrom 1 benchmark result across 1 source* thinly tested, treat the rank as provisional

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 ABR asr-19m-v2-en, out of 176 with a score.

#modelscorewhereRAMreleased
147GigaAM Multilingual (CTC 220M) Sber42±7local~0.9GB2026-07
148ESPnet Transformer 18-layer + RNN LM ESPnet42±4local~1GB2020-07
149ABR asr-19m-v2-en Applied Brain Research41±5local~0.1GB2025-11
150SpeechBrain Transformer + TransformerLM (LibriSpeech) SpeechBrain40±7comparison only2022-03
151ESPnet Multi-blank Transducer 8421 ESPnet40±4local~8GB2023-02

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

$ what it's good at

other tests1 not in a task · 90% of its score
testresultvs the others in its rundifficultycountswhy
Open ASR English short-form WER (March 2026)10.6% ✎▁▁▁▃··▁▁▃██▆ beats 18/86avg 8.2%▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 90%blunt

$ where to run it

Other ways to run it

  1. Run it as published · needs ~0.08 GB · from huggingface.co/abr-ai
    uvx --from huggingface_hub hf download abr-ai/niagara-19m-batch.en --revision dab6545337495482f2fc05455432a7a05c88d3cc --local-dir asr-19m-v2-en && uv run --with torch --with transformers --with sentencepiece --with soundfile python -c "import soundfile as sf; from transformers import AutoFeatureExtractor, AutoModel, AutoTokenizer; p = './asr-19m-v2-en'; feature_extractor = AutoFeatureExtractor.from_pretrained(p, trust_remote_code=True); model = AutoModel.from_pretrained(p, trust_remote_code=True); tokenizer = AutoTokenizer.from_pretrained(p, trust_remote_code=True); audio, _ = sf.read('audio.wav', dtype='float32'); print(tokenizer.decode_from_logits(model(feature_extractor(audio)))[0])"
$ how the score was worked out · Open ASR Leaderboard 90%

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▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 90%Apache-2.0
prior · pull toward the middle▇▇▇ 10%
weighted average= 40.8

$ 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