# microsoft/VibeVoice

VibeVoice: Open\-Source Frontier Voice AI

- HackerByte page: <https://hackerbyte.net/en/projects/vibevoice-open-source-frontier-voice-ai-1044296738>
- GitHub: <https://github.com/microsoft/VibeVoice>
- Programming language: Python

Open\-Source Frontier Voice AI

## AI-generated summary

VibeVoice is a family of open\-source AI models from Microsoft focused on speech processing. It includes advanced Automatic Speech Recognition \(ASR\) capable of handling long\-form audio with speaker diarization and text\-to\-Speech \(TTS\) for generating extended, multi\-speaker, and expressive speech. The models leverage efficient continuous speech tokenizers and diffusion frameworks for high\-fidelity audio synthesis and transcription.

## Key points

- VibeVoice\-ASR handles up to 60 minutes of audio in a single pass, providing speaker diarization and timestamps.
- VibeVoice\-TTS can synthesize up to 90 minutes of speech with support for up to 4 distinct speakers.
- Real\-time streaming TTS is available with VibeVoice\-Realtime\-0.5B, featuring low latency and a small parameter size.
- Models support multiple languages, with ASR covering over 50 languages and TTS offering multilingual voices.
- Advanced features include customized hotwords for ASR and expressive speech generation for TTS.