parlor

Why we included this project

Parlor is an early research preview that shows what a fully local voice assistant can do: it carries on real-time spoken conversations, complete with vision and streaming speech, without sending audio to any cloud service. The stack runs speech recognition, a locally hosted model that generates replies, and sentence-by-sentence text-to-speech, with a fast turn-taking heuristic that decides when you have finished talking. The author built it while self-hosting a free English tutor, and the project is upfront about being rough around the edges. Developers who want to see how to wire together websocket audio streaming, model serving, and low-latency turn detection will find the architecture laid out clearly enough to copy.