VibeVoice-ASR
by Microsoft · Model · Voice & audio
Supports over 50 languages natively; finetuning code and vLLM inference are available
AccessOpen weightsNeeds~1.6 GBDownload1.6 GBSize7B parametersLicenseMIT
Evidence
- Jul 23VibeVoice-ASR-BitNet for CPU: heterogeneous quantization shrinks 4.62 GB to 1.58 GB, real-time (RTF < 1) on 3+ CPU threads without a GPU
- Mar 12Available to test in Azure AI Foundry Labs
- Jan 21Open-sourced 7B speech-to-text handling 60 minutes of audio in one pass, outputting speaker, timestamps and content with custom hotwords