Unmute

Unmute gives text LLMs a spoken conversation interface using Kyutai speech recognition and speech synthesis. It's for developers who want to self-host voice chat while choosing their own language model.

• Spoken responses: The system transcribes your speech and starts reading the model's reply while it’s still generating text, with speech models optimized for low latency.
• Local or external models: Connect OpenAI-compatible LLM servers, including local Ollama and vLLM instances or external services such as OpenRouter.
• Voices and characters: Customize character voices and prompts, and show subtitles for both the user and the chatbot.

Unmute is open source under the MIT license. It supports Docker and deployment without Docker; the default Docker setup runs the speech services and vLLM locally. With local model services, conversation processing can stay on your hardware. Using an external LLM sends the text exchange to that provider, while the hosted browser demo at Unmute.sh processes conversations on remote servers.

Self-hosting requires an x86_64 machine and a CUDA-capable GPU with at least 16 GB of VRAM. Running speech recognition, speech synthesis and the LLM on separate GPUs can reduce latency compared with a single-GPU setup. Unmute doesn't provide built-in tool calling.