Unmute
A system that gives a text-based LLM a voice by wrapping it with low-latency speech-to-text and text-to-speech models, so a user can talk to it out loud.
Unmute lets any text-based LLM listen and speak. A speech-to-text model transcribes what the user says, the LLM generates a text response, and a text-to-speech model reads that response aloud, with the backend coordinating a frontend, an STT server, an LLM server and a TTS server over WebSockets in real time. The STT and TTS components are Kyutai's own models, optimized for low latency.
The LLM can be served through OpenRouter or self-hosted with vLLM. Deployment requires a Linux machine, or Windows with WSL, and a CUDA GPU with at least 16 GB of VRAM; the project ships Docker Compose, Dockerless and Docker Swarm deployment paths.