Real-Time AI Voice Chat

Real-Time AI Voice Chat is a client-server application for holding low-latency spoken conversations with large language models. The browser captures microphone audio and streams it over WebSockets to a Python backend. RealtimeSTT converts speech into text, which is sent to an LLM such as Ollama or OpenAI for response generation. RealtimeTTS then synthesizes the answer and streams speech back to the browser. Dynamic silence detection improves turn taking, while interruption handling lets users speak over an ongoing response. Multiple speech engines, including Kokoro, Coqui, and Orpheus, can be selected. Docker Compose simplifies deployment, although the original maintainer now treats the project as community-driven rather than actively developed.