whatmodel.sh / stt / LiteWhisper large-v3-acc

LiteWhisper large-v3-acc

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

LiteWhisper large-v3-acc is an open-weights speech-to-text model from Efficient-Speech (LiteASR), released February 2025. Its published size is 1.4B parameters. Licence: apache-2.0.

Catalogue identity notes

Exact checkpoint efficient-speech/lite-whisper-large-v3-acc (low-rank compressed Whisper encoder; custom code, trust_remote_code). Params: HF safetensors total. released: HF repo creation month. Command follows the card snippet (processor from openai/whisper-large-v3).

55.0±3.1#71 of 162 speech-to-text modelsfrom 4 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 LiteWhisper large-v3-acc, out of 176 with a score, then more from Efficient-Speech (LiteASR).

#modelscorewhereRAMreleased
80Fun-ASR-Nano-2512 Alibaba (FunAudioLLM)55±3local~0.9GB2025-12
81Scribe v2 Realtime ElevenLabs55±3api $0.0065/audio min2025-11
82LiteWhisper large-v3-acc Efficient-Speech (LiteASR)55±3local~3.9GB2025-02
83Omnilingual ASR CTC 3B v2 Meta55±3local~7.9GB2025-12
84Omnilingual ASR CTC 7B Meta55±2local~16GB2025-11
more from Efficient-Speech (LiteASR)
79LiteWhisper large-v3 Efficient-Speech (LiteASR)55±3local~3.7GB2025-02
95LiteWhisper large-v3-turbo-acc Efficient-Speech (LiteASR)53±4local~2.1GB2025-02
101LiteWhisper large-v3-fast Efficient-Speech (LiteASR)52±3local~3.6GB2025-02

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

$ what it's good at

other tests4 not in a task · 96% of its score
testresultvs the others in its rundifficultycountswhy
Open ASR English short-form WER (March 2026)7.2% ✎▁▁▁▃··▁▁▂██▆ beats 53/86avg 8.2%▇▇▇▇▇▇▇▇▇▇▇▇▇ 33%blunt
Multilingual LibriSpeech French test WER lab7.4% ✎█·······██·█ beats 1/3avg 11%▇▇▇▇▇▇▇▇▇▇ 25%typical
LiteASR paper · ESB 8-set average WER lab10.1% ✎▂·······▂··█ beats 5/6avg 13%▇▇▇▇▇▇▇▇▇ 22%typical
Multilingual LibriSpeech German test WER lab8.7% ✎█····█···█·█ beats 2/3avg 13%▇▇▇▇▇▇▇ 17%typical

$ where to run it

Other ways to run it

  1. Run it as published · needs ~3.86 GB, NVIDIA only · from huggingface.co/efficient-speech (official)
    uv run --with transformers --with torch --with librosa python -c "import librosa, torch; from transformers import AutoProcessor, AutoModel; m=AutoModel.from_pretrained('efficient-speech/lite-whisper-large-v3-acc', trust_remote_code=True).to(torch.float16).to('cuda'); p=AutoProcessor.from_pretrained('openai/whisper-large-v3'); f=p(librosa.load('audio.wav', sr=16000)[0], sampling_rate=16000, return_tensors='pt').input_features.to(torch.float16).to('cuda'); print(p.batch_decode(m.generate(f), skip_special_tokens=True)[0])"

Official weights and documentation: huggingface.co/efficient-speech ↗

$ how the score was worked out · LiteASR (arXiv:2502.20583v2)… 64%, Open ASR Leaderboard 33%

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
LiteASR (arXiv:2502.20583v2) · ESB WER of Whisper and LiteASR-compressed encoders · lab-reported3▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 64%CC BY 4.0
Open ASR Leaderboard1▇▇▇▇▇▇▇▇▇▇ 33%Apache-2.0
prior · pull toward the middle▇ 4%
weighted average= 55.0

$ 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