All models
realtime

Kyutai TTS

CC-BY-4.0

Word-level streaming TTS built for live voice agents

Kyutai TTS pairs a 1B backbone with a 600M depth transformer to stream audio at the word level, generating 12.5Hz frames fast enough for real-time conversational agents. Voice cloning is deliberately disabled — it only speaks from a fixed set of precomputed voice embeddings, a safety choice that trades flexibility for guardrails against impersonation.

Hear Kyutai TTS

Standardized samples are pending for this release. See the official sources below for the author’s demo.

Kyutai TTS · Reference-based · Same three scripts across the directory.

Best for: Word-level streaming TTS built for live voice agents

Before you choose: Compare the samples with your own use case. Check the linked documentation for code, weight, and voice license terms.

Details awaiting a fresh review. Hardware figures are estimates unless a benchmark is linked.

No samples yet

Every model in this directory is read against the same three scripts so voices can be compared honestly — Kyutai TTS’s samples just haven’t been generated yet.

Contribute samples on GitHub →

Run Kyutai TTS

pip install moshi

Sources & setup details

Kyutai TTS FAQ

Do I need a GPU to run Kyutai TTS?

The directory lists roughly 8 GB of VRAM as a historical estimate. Check the model documentation for your checkpoint and runtime.

Can Kyutai TTS clone voices?

No — Kyutai TTS uses preset voices. Browse the voice-cloning category for models that clone from reference audio.

What license is Kyutai TTS released under?

Kyutai TTS is released under the CC-BY-4.0 license. Always check the repository for the exact terms — some models license code and weights separately.

What languages does Kyutai TTS support?

Kyutai TTS supports English, French.

Is there a hosted Kyutai TTS API?

OpenSpeech Cloud is accepting production-access interest for Kyutai TTS, but this model is not hosted in the beta. Use the official repository for current deployment options.