Moshi is a speech-text foundation model and full-duplex spoken dialogue framework. It uses Mimi, a state-of-the-art streaming neural audio codec.
kyutai-labs/moshi is #2,256 on the bestofgithub Elo ladder, in Platinum II, with a rating of 1,700 (+28 this week). It has 11,530 stars, 73 gained in the last week.