Voice AI
Real-time speaker diarization on a single mic, explained.
One microphone. Two speakers. Zero training data per call. How diart, pyannote, and a 240ms window do the work.
Dani ParkMay 9, 2026
Engineering, customer stories, and voice-AI essays from the team building Vought. Written by the operators who ship it.
One microphone. Two speakers. Zero training data per call. How diart, pyannote, and a 240ms window do the work.
The AI is invisible. So how do you animate it? Four signature motions, 240ms each, and a discipline about negative space.
The engineering deep-dives, customer numbers, and voice-AI essays we publish here — delivered once a month. Unsubscribe inline.
· 412 subscribers · one essay every fourth Tuesday.