Show HN: RealtimeVAD – trading false positives for 110ms earlier end-of-speech
Summary
RealtimeVAD introduces faster end-of-speech signaling by combining Silero VAD with a second, faster check that analyzes room noise and speaker level every 10 ms. The approach yields earlier end-of-speech events and more frequent, smaller resets, with benchmarks across thousands of turns. The project provides a quick start, live demo, and benchmarks, and is MIT-licensed.