Kyutai Releases 1.6B Bilingual TTS Model

French AI research lab Kyutai has released a new 1.6-billion-parameter text-to-speech (TTS) model that generates high-quality audio in both English and French. The model is available under a permissive Creative Commons license (CC-BY-4.0), allowing for broad use, including in commercial applications.

The system is built on what the team calls the Moshi stack, an adaptation of the VoiceCraft architecture. A key feature is its ability to support streaming audio generation, making it suitable for real-time applications where low latency is critical, such as conversational agents or live narration tools.

A New Voice in Open TTS

While the open-source landscape for large language models has exploded, the space for high-quality, permissively licensed TTS models is less crowded. Kyutai's release provides a significant new building block for developers creating voice-enabled products without relying on proprietary, closed-source APIs.

As a non-profit lab, Kyutai's contribution adds valuable diversity to an ecosystem often dominated by a few large tech companies. Developers can access the model weights, explore an interactive demo, and find usage examples on the official Hugging Face repository.