High Voice & Audio · 1 min read
Moshi: Kyutai's first open-source full-duplex voice assistant
In one sentence French non-profit lab Kyutai unveils Moshi, a full-duplex voice assistant with ~200ms latency based on a single multimodal model handling simultaneous input and output audio.
Reading level
Kyutai, a newly founded French non-profit AI lab, demos a voice assistant that talks in real time. Its name is Moshi.
Key difference vs Siri/Alexa: normally you speak, pause, the assistant replies. Moshi can listen and speak at the same time, like a real person who interrupts or backchannels while you're still talking.
Live demo in front of the press, fully open source: model, code, weights. First time something like this is publicly available.
Companies
Kyutai
Tools
Moshi
Tags
KyutaiMoshiVoiceReal-timeOpen SourceStreaming
Sources