OpenVoiceChat
Open-source framework for conversational voice AI. 2023–present.
Why
Most voice assistants are either closed products or a pile of separate STT, TTS, and LLM pieces you have to wire together yourself. I wanted the whole loop — listening, thinking, speaking, interrupting — to feel like one system.
What
OpenVoiceChat is an open-source framework for building conversational voice agents. It connects speech recognition, a language model, and speech synthesis, with turn-taking and interruption handling.
Hard part
Latency and turn-taking. A correct answer that arrives late feels broken. The system has to detect when the user has finished, handle interruptions, and keep the pipeline moving without making the conversation feel robotic.
What I learned
I wrote about the build and the lessons in 200 Commits Later.
Links
- GitHub — fakhirali/OpenVoiceChat
- Demo — fakhirali.github.io/OpenVoiceChat
- Write-up — 200 Commits Later