Open-source voice AI just reached a new milestone. Two cutting-edge models are now available:

FireRedTTS2 delivers impressive performance metrics—140ms latency with support for 4-speaker dialogue interactions across 7 languages. Built on a dual-transformer architecture, it handles complex audio processing while maintaining real-time responsiveness.

VibeVoice takes conversation length to another level, supporting 90-minute continuous interactions with genuine real-time processing capabilities. The architecture enables natural, extended dialogues without degradation.

Both models represent significant steps forward in open-source voice AI development, combining low-latency performance with practical multi-language and multi-speaker capabilities.