whatmodel.sh / stt / VibeVoice-ASR-Streaming-7B

VibeVoice-ASR-Streaming-7B

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

VibeVoice-ASR-Streaming-7B is an open-weights speech-to-text model from Microsoft, released September 2026. Its published size is 8.67B parameters. Licence: mit.

Catalogue identity notes

Exact checkpoint microsoft/VibeVoice-ASR-Streaming-7B @ 60d858b (MIT, created 2026-09-02): the released 22-frame (2.9 s) chunk, 4-frame lookahead configuration ('We release 1.5B and 7B model weights for the 22-frame (2.9-second) chunk configuration', arXiv:2609.02812v2). Distinct from the non-streaming VibeVoice-ASR / vibevoice-asr-hf and the 1.5B streaming model.

Model information sources

Release date ↗ · Weights licence ↗ · Parameter count ↗

HF safetensors metadata total 8,674,021,857 (BF16), incl. the frozen acoustic/semantic tokenizer encoders; the maker's '7B' refers to the Qwen2.5-7B backbone.

47.2±5.1#112 of 162 speech-to-text modelsfrom 3 benchmark results across 1 source

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-Streaming-7B, out of 176 with a score, then more from Microsoft.

#modelscorewhereRAMreleased
123Distil-Whisper distil-small.en Hugging Face (Distil-Whisper)48±4local~0.9GB2023-12
124CrisperWhisper Nyra Health47±3local~4.4GB2024-08
125VibeVoice-ASR-Streaming-7B Microsoft47±5local~17GB2026-09
126SpeechBrain Conformer (LargeScaleASR / Loquacious) SpeechBrain47±5local~2.8GB2025-02
127GigaAM Multilingual Large (CTC 600M) Sber47±7local~2.3GB2026-07
more from Microsoft
57Phi-4-multimodal-instruct Microsoft58±2local~14GB2025-02
102VibeVoice-ASR-HF Microsoft52±3local~21GB2026-03
109VibeVoice-ASR-BitNet Microsoft50±4local~1.7GB2026-07

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

$ what it's good at

other tests3 not in a task · 90% of its score
testresultvs the others in its rundifficultycountswhy
GigaSpeech test WER lab10.2% ✎▁····▁··▁▄█▂ beats 6/17avg 10.0%▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 40%small run
LibriSpeech test-other WER lab6.5% ✎▁▁▁▂▅▃▅▄█▂▂▁ beats 11/54avg 4.8%▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 38%small run
LibriSpeech test-clean WER lab2.3% ✎▁·▁·▁▁▁▁▃▇█▆ beats 39/100avg 2.1%▇▇▇▇▇ 12%blunt · small run

$ where to run it

Other ways to run it

  1. Run it as published · needs ~17.36 GB, NVIDIA only · from github.com/microsoft (official)
    git clone https://github.com/microsoft/VibeVoice.git && cd VibeVoice && pip install -e . && python demo/vibevoice_asr_streaming_inference_from_file.py --model_path microsoft/VibeVoice-ASR-Streaming-7B --audio_files audio.wav

Official weights and documentation: huggingface.co/microsoft ↗

$ how the score was worked out · VibeVoice-ASR-Streaming Techn… 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
VibeVoice-ASR-Streaming Technical Report (arXiv:2609.02812v2), Table 3 single-speaker recognition · lab-reported3▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 90%CC BY 4.0
prior · pull toward the middle▇▇▇ 10%
weighted average= 47.2

$ 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