TL;DR — In a browser voice app, "did the candidate stop talking? " is a real-time signal you cannot reliably get from speech-to-text. STT event timings are polluted by network latency, buffering, and the recognizer rewriting its own guesses, so they will report a pause while someone is mid-sente...
Source: [Dev.to](https://dev.to/ji_ai/why-i-measure-interview-silence-with-webaudio-not-the-speech-api-4gmh)