VibeVoice ASR HF
Microsoft · released Mar 2, 2026 · microsoft/VibeVoice-ASR-HF
Speech to textTranscribes a recording into words
- Type
- Open weightsMIT License
- Languages
- 51
- Size
- 8.3B
Context measured in tokens
Our take
Written Sep 4, 2026VibeVoice ASR HF is Microsoft's MIT-licensed speech-to-text model with 8.3 billion parameters and support for 51 languages. It transcribes podcasts and video faster and more accurately than most, but struggles with clean read-aloud audio, meetings and accented speech.
Pick this for podcast or video transcription where speed matters — it processes an hour of audio in about sixteen seconds and beats most rivals on that condition. Use it for multilingual pipelines needing broad language coverage with a permissive licence. Skip it if you need clean-audio accuracy, meeting transcription, or heavily accented speech, or if you want a hosted provider rather than self-hosting.
The case for it
- Better than most on podcast and video audio, with a 7.9% word error rate against an 8.2% field median.
- Very fast batch processing at 221 times real time on benchmark hardware.
- 51 languages under an MIT licence, allowing commercial use, modification and redistribution.
- Strong on European-accented parliamentary speech at 2.0% word error rate.
The case against it
- Worse than most on clean read-aloud audio and meeting-room audio with crosstalk, at 1.6% and 12.0% word error rates against 1.4% and 10.6% medians.
- Worse than most on accented speech from diverse international speakers, at 12.2% against a 10.8% median.
- No commercial hosting options available; you must self-host.
How good is it?
TranscriptionTurning speech into text3 of 5Open ASR WER · 50th of 76
94.4%
Misses roughly one word in 18, averaged over nine English test sets.
219×56th of 74
an hour of audio in 16 seconds, on the board's own hardware. Your machine will differ.
51
Listed on the model card. The accuracy above is English only.
Percentage of words wrong on each set, lower better. Bars are scaled to this model's own worst case; the placing beneath each rate is against every model measured on that set.
Which languages ↓ ↑
English · Chinese · Spanish · Portuguese · German · Japanese · Korean · French · Russian · Indonesian · Swedish · Italian · Hebrew · Dutch · Polish · Norwegian · Turkish · Thai · Arabic · Hungarian · Catalan · Czech · Danish · Persian · Afrikaans · Hindi · Finnish · Estonian · aa · Greek · Romanian · Vietnamese · Bulgarian · Icelandic · Slovenian · Slovak · Lithuanian · Swahili · Ukrainian · kl · Latvian · Croatian · Nepali · Serbian · Filipino · Yiddish · Malay · Urdu · Mongolian · Armenian · Javanese
The figures above come from the Open ASR Leaderboard, an independent public test that runs every model on the same recordings. It is the only measurement of transcription quality we know of, so there are no other scores to show.
Each of these is the same transcription job on a different kind of recording, so together they say where it holds up and where it slips — not how closely it follows an instruction.
Every published score for this model9 scoresEvery figure we hold, from 9 boards, with who ran it and a link to the source — including the boards no rating above is built on.
Can you run it yourself?
Comfortable fit
GeForce RTX 4090 · 24 GB
Room to spare. 16.1 GB spare means a 10% error in the size would not change the answer.
GeForce RTX 5090 · 32 GB
Room to spare. 24.1 GB spare means a 10% error in the size would not change the answer.
Apple M1 (8-core GPU) · 16 GB
Room to spare. 5.3 GB spare means a 10% error in the size would not change the answer.
These cards answer whether VibeVoice ASR HF loads, not how fast it transcribes. Throughput figures for a transcription model come from its text decoder, so treat this as a fit answer rather than a speed one.
Memory use by level
Against a 24 GB card.
What is quantisation? →This model on every device we track71 devicesThe Q4 build most people download, on each device: what the weights come to, how much context the memory leaves, and whether it runs. Smallest device that runs it first. This is a fit answer: whether it loads, not how fast it transcribes.
When we formed this view
Recent changes
What moved
first indexed by our pipelineEach date is the day we first saw the change, or the day the maker announced it.
What we do not know about this model yet
- We hold no measured file for it, so all 3 sizes on this page are calculated from the parameter count.
- Nothing we hold says whether an endpoint streams, so we do not show it either way.
Licence and identifiers
What the licence allowsMIT License, what it allows commercially, and the identifiers you need to pull this model — its Hugging Face repo, our slug and a machine-readable card.
Licence
MIT License
Fully permissive: do anything with attribution. No patent grant, unlike Apache-2.0.
Identifiers
- Hugging Face
- microsoft/VibeVoice-ASR-HF
- Architecture
- Dense
- Takes in, gives back
- Audio in, text out
- Catalogue slug
- microsoft-vibevoice-asr-hf