How voice AI hears: from 16,000 samples per second to spectrograms
prdeepakbabu · x · 2026-10-18
Developer prdeepakbabu built interactive teaching aids showing what speech-recognition systems actually "hear." A mic converts air-pressure changes into numbers—typically 16,000 samples per second for speech AI. A short-time Fourier transform slices these into 25 ms overlapping frames and measures energy per frequency, producing a spectrogram: a whistle is one bright line, a clap a vertical stripe across many frequencies, a hiss is high-frequency noise, and vowels show vocal-fold harmonics plus formants that distinguish "ee" from "ah." He'll present this at IEEE STEM Summit 2026.
Related event: Open-source browser tool visualizes how speech AI hears sound(2 posts)→
More from Companies & People
- Sentry CEO: Dots email experiment vanished from app but keeps sending notifications — zeeg · 2026-10-18
- Nathan Lambert questions Kimi serving Claude tokens directly to users — natolambert · 2026-10-18
- Anthropic got highest junk credit rating in June ahead of $2T IPO target — GaryMarcus · 2026-10-18
- AI Now hosts "Building Worker Power in the Era of AI" discussion on Oct 19 — AINowInstitute · 2026-10-18
- Stripe CEO Patrick Collison's advice to young ambitious people: go deep, read a lot, make things — benaratame · 2026-10-18
- Armin Ronacher rebuts Jung quote: "You must work hard and seek distribution" — draecomino · 2026-10-18