VibeVoice

VibeVoice is Microsoft Research's open-source frontier voice AI family, delivering multi-speaker text-to-speech and speech recognition built on ultra-low-frame-rate (7.5 Hz) continuous speech tokenizers. Ships three MIT-licensed checkpoints: a 7B ASR model covering 50+ languages with 60-minute clips, a 1.5B TTS model generating up to 90 minutes of audio with 4 speakers, and a 0.5B real-time model for 9 languages. Deployable via Hugging Face, Google Colab, or vLLM — aimed at researchers and developers needing long-form, multi-voice synthesis for podcasts and conversational audio.