whatmodel.sh / stt / VibeVoice-ASR-HF

VibeVoice-ASR-HF

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

VibeVoice-ASR-HF is an open-weights speech-to-text model from Microsoft, released March 2026. Its published size is 8.33B parameters. Licence: mit.

Catalogue identity notes

Exact checkpoint microsoft/VibeVoice-ASR-HF (Transformers-native release, transformers>=5.3); not the TTS model vibevoice-1-5b. Params: HF safetensors total. released: HF repo creation month.

51.8±3.2#91 of 162 speech-to-text modelsfrom 6 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 VibeVoice-ASR-HF, out of 176 with a score, then more from Microsoft.

#modelscorewhereRAMreleased
100Rev AI Machine (Reverb ASR) Rev52±3api $0.0033/audio min
101LiteWhisper large-v3-fast Efficient-Speech (LiteASR)52±3local~3.6GB2025-02
102VibeVoice-ASR-HF Microsoft52±3local~21GB2026-03
103OWSM-CTC v4 1B ESPnet (CMU WAVLab)52±4local~5.4GB2025-01
104Omnilingual ASR CTC 1B Meta51±2local~2.9GB2025-11
more from Microsoft
57Phi-4-multimodal-instruct Microsoft58±2local~14GB2025-02
109VibeVoice-ASR-BitNet Microsoft50±4local~1.7GB2026-07
125VibeVoice-ASR-Streaming-7B Microsoft47±5local~17GB2026-09

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

$ what it's good at

other tests6 not in a task · 96% of its score
testresultvs the others in its rundifficultycountswhy
Open ASR English short-form WER (March 2026)7.8% ✎▁▁▁▃··▁▁▂██▆ beats 43/86avg 8.2%▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 35%blunt
LibriSpeech test-other WER lab5.8% ✎███·····██·█ beats 2/5avg 5.7%▇▇▇▇▇▇▇▇▇▇▇ 27%typical
VoxPopuli English test WER lab4.9% ✎▄··▄··▄···█▄ beats 5/5avg 7.4%▇▇▇▇▇ 12%typical
LibriSpeech test-clean WER lab2.2% ✎█·█·█·█···██ beats 2/5avg 2.2%▇▇▇ 7.8%blunt
FLEURS Mandarin test error rate lab7.9% ✎▄·▂▄▂▄██▂▂·▆ beats 14/20avg 7.3%▇▇▇ 7.4%typical
FLEURS English test WER lab4.7% ✎▄·····▄▄▄··█ beats 3/5avg 5.5%▇▇▇ 7.0%blunt

$ where to run it

Other ways to run it

  1. Run it as published · needs ~20.49 GB, NVIDIA only · from huggingface.co/microsoft (official)
    uv run --with transformers --with torch --with accelerate python -c "from transformers import AutoProcessor, VibeVoiceAsrForConditionalGeneration as M; r='microsoft/VibeVoice-ASR-HF'; p=AutoProcessor.from_pretrained(r); m=M.from_pretrained(r, device_map='auto'); i=p.apply_transcription_request(audio='audio.wav').to(m.device, m.dtype); o=m.generate(**i); print(p.decode(o[:, i['input_ids'].shape[1]:], return_format='transcription_only')[0])"

Official weights and documentation: huggingface.co/microsoft ↗

$ how the score was worked out · microsoft/VibeVoice-ASR-BitNe… 61%, Open ASR Leaderboard 35%

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
microsoft/VibeVoice-ASR-BitNet model card (Accuracy table) · lab-reported5▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 61%MIT
Open ASR Leaderboard1▇▇▇▇▇▇▇▇▇▇▇ 35%Apache-2.0
prior · pull toward the middle▇ 4%
weighted average= 51.8

$ 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