VibeVoice-ASR-BitNet
VibeVoice-ASR-BitNet is Microsoft Research's compressed automatic speech recognition model for real-time CPU and edge transcription without a GPU.
Provider
Microsoft Research
Model family
Microsoft VibeVoice
Speech recognition model
Cost tier
Asr Bitnet
Status
Current
Release Jul 24, 2026
Why teams choose it
Evaluate on the target CPU, thread count, microphone conditions, accents, and domain
vocabulary before accepting real-time claims.
This is the Microsoft candidate to track when private or edge transcription matters
more than maximum hosted-model accuracy.
Tradeoffs to know
- The model card reports seven languages, so unsupported-language behavior needs separate evaluation.
- Published speed comparisons depend on hardware, threads, build options, and audio conditions.
Benchmarks
No independent or vendor benchmark scores in this catalog entry.
Technical specs
- Inputs
- audio
- Outputs
- text
- License
- MIT
- Languages
- 7 (official model card)
- Model Size
- 1.58 GB (official model card)
- Capabilities
- speech recognition
- CPU inference
- edge deployment
- multilingual transcription
- low-memory serving
Related models
VibeVoice-ASR-BitNet FAQ
What is VibeVoice-ASR-BitNet?
VibeVoice-ASR-BitNet is Microsoft Research's compressed automatic speech recognition model for real-time CPU and edge transcription without a GPU. Its official model card documents a 1.58 GB footprint, seven-language support, and real-time CPU performance under the tested configuration.
When does VibeVoice-ASR-BitNet fit best?
On-device transcription
What should teams watch out for with VibeVoice-ASR-BitNet?
The model card reports seven languages, so unsupported-language behavior needs separate evaluation.
Explore next
Models, tools, and comparisons that connect to this reference.