VIBEVOICE

Summary from the TLDR Dev newsletter

VibeVoice is an open-source frontier voice AI project by Microsoft that provides models for Text-to-Speech and Automatic Speech Recognition, using continuous speech tokenizers and a next-token diffusion framework with LLMs for high-fidelity processing of long audio. The project features VibeVoice-ASR for multilingual recognition and VibeVoice-TTS for multi-speaker and streaming synthesis.