One Voice Detector to Rule Them All
thegradient.pub
thegradient.pub
You can add some basic spectral analysis and make it even better, with minimal latency implications. "Cepstrum" and "spectral entropy" are some valuable search terms to apply to this.
It was good enough to demo, though.
(for those who don't recall, the Nest Smoke Alarm had voice deactivation that didn't really work well. https://bgr.com/general/google-nest-smoke-alarm-video/)
I'd like to break up audio into sentences/lyrics and that is exactly what I need for that.