Speech To Speech
Local voice agent pipeline chaining VAD, speech recognition, LLM and TTS
What it is
A Python package that runs a spoken conversation loop on your own computer: voice activity detection, transcription, a language model reply, and synthesized speech, with each stage swappable for other open models. Its server implements the OpenAI Realtime protocol, reachable over WebSocket or WebRTC, which lets existing Realtime clients, the OpenAI Agents SDK, or LiveKit Agents connect to a local machine instead. Developers building voice assistants, robots, or avatars use it to keep audio and models offline without API keys.