RealtimeVAD: end-of-speech detection about 110ms earlier for voice assistants
A CPU-only, MIT-licensed drop-in for Silero VAD that adds a 10ms energy check with room-noise tracking, detecting end of speech a median 110ms earlier on about 3,200 real turns — at the cost of roughly 10x more false positives.
Why it matters
Trades false positives for speed: the earlier end-of-speech signal only pays off in pipelines that can discard speculative work instantly, which the author built it for.
What you could build with it
A developer building a local voice assistant could shave about a tenth of a second off the pause before the assistant answers, making conversations feel snappier on a small home computer with no cloud round trip.
Does it hold up?
Too early to judge — one author's benchmark on his own voice data; the tradeoff only works in pipelines that can drop speculative generations the moment the user keeps talking.
Built with RealtimeVAD
- Show HN: RealtimeVAD – trading false positives for 110ms earlier end-of-speechhackernews · Launch discussion: the author reports a 92ms gain reproduced by another user on noisier recordings, and that false positives run near 90% on speculative generations.
- Trading false positives for ~110ms earlier end-of-speech in voice pipelines: RealtimeVADvibingtalk.com · The maker's writeup: median 110ms earlier than Silero on about 3,200 real turns, at the cost of roughly 10x more false positives on breaths and pauses.
- Show HN: RealtimeVAD - trading false positives for 110ms earlier end-of-speechainews247.org · News summary of the Show HN launch: CPU-only, MIT, a drop-in for Silero's VAD iterator, aimed at pipelines that can discard speculative work instantly.
Learn more
First spotted on github: source.
More AI models
VoiceStudio: open-source, fully-local ElevenLabs alternative for voice cloning, dubbing, and audiobooksdebpalash/VoiceStudio is a desktop app that runs voice cloning, voice design, video dubbing, dictation, transcription…models · JEV 0.75Edge0-145B: open 145B MoE that runs on a single consumer GPUEdge0 open-sourced 145B and 120B latent-MoE models derived from Qwen3.8-Flash-Next, plus an inference framework that…models · JEV 0.74Strata: run a 125B model on a gaming PCFree, MIT-licensed local inference engine that runs Qwen3.8-Flash-Next (125B MoE, about 6B active per token) on…models · JEV 0.73Google AI Edge Gallery: offline speech recognition and synthesis on-deviceGoogle added fully offline speech recognition and text-to-speech to AI Edge Gallery, with ML Kit APIs on Android and…models · JEV 0.73Soup: fine-tune an 8B LLM on a 4 GB laptop GPU from one YAML fileAn open-source CLI that fine-tunes LLMs from a single YAML file, with layer streaming that trains an 8B model on a 4…models · JEV 0.73Tencent Hunyuan Hy3: open 295B MoE with 21B active under Apache 2.0Tencent open-sourced Hy3, a 295B-parameter mixture-of-experts model with only 21B parameters active per token, under…models · JEV 0.73
Get the week's best AI launches, plus 3 ideas worth building
One email every Saturday. Ranked by traction, not hype. Free.