Show HN: RealtimeVAD – trading false positives for 110ms earlier end-of-speech

RealtimeVAD is a voice activity detection system that identifies when a speaker has stopped talking approximately 100ms faster than the standard Silero VAD by combining Silero's speech detection with a faster real-time audio level comparison. The trade-off is increased false positives in short gaps between words, occurring about ten times more frequently than Silero alone. The tool runs efficiently on a single CPU thread and requires only numpy and onnxruntime, making it suitable for real-time voice assistant applications.