MultimodalAlexandre Défossez et al. (Kyutai)
Kyutai unveils Moshi
Moshi presented — a full-duplex speech-to-speech model that does away with the VAD, ASR and TTS pipeline. Theoretical latency 160 ms, measured 200 ms. Inner Monologue: time-aligned text tokens are generated before the audio tokens
Why it matters
The lineage that replaced pipeline-based voice dialogue with a single model
Source — Primary sources first; secondary reports are labelled “Press”
Paperarxiv.orghttps://arxiv.org/abs/2410.00037Added in Week 35, 2026 · See that week