Pronunciation of the day: Moshi

Kyutai's real-time full-duplex speech-text foundation model (July 2024), built on the Mimi neural audio codec (Mimi = MEE-mee). Say it MOH-shee: stress the first syllable, long 'o' as in 'go', 'sh' as in 'ship' — MOH-shee. The name matches the Japanese word もしもし (moshi moshi), the phrase used to answer the phone; that etymology is widely assumed but is NOT stated in Kyutai's README, the arXiv paper (2410.00037), or kyutai.org, so treat the naming rationale as unconfirmed. Common English mispronunciations are 'MOSH-ee' (rhyming with 'posh') and 'MOSH-eye' — both wrong.

For the full source citation, alternate readings, and CLI usage, see the canonical page → Moshi.