tools
Unmute
Open-source software that gives text-based language models real-time speech input and output.
Unmute connects speech-to-text, a text language model, and text-to-speech. It transcribes speech in real time, detects when a speaker has finished, sends text to an LLM, and begins speaking before the response is complete.
Developers can use the demo or self-host the system with Docker Compose or other deployment options. It supports external OpenAI-compatible LLM servers and requires suitable GPU hardware for self-hosting. The project does not preserve emotion, emphasis, hesitation, or tone that the LLM receives only as text.
Features
- Real-time speech-to-text transcription
- Semantic voice activity detection for turn-taking
- Works with any text language model
- Streaming text-to-speech before the full response is generated
- Function-calling and external-tool integrations
- Self-hosting with Docker Compose, Dockerless, or Docker Swarm
- Open-source under the MIT license
- Supports OpenAI-compatible external LLM servers
Use cases
- Build voice assistants with a text language model
- Create voice interfaces for accessibility tools
- Prototype lower-latency customer support hotlines
- Develop voice-based programming tools
- Run spoken role-playing characters for games