Dev Tools · 16h ago
Measuring Interview Silence with WebAudio, Not Speech API
A developer building a voice mock interviewer found that the Web Speech API's timing is unreliable for detecting silence due to network latency and buffering. Instead, they use WebAudio to measure acoustic silence directly, with hysteresis and dual thresholds to distinguish thinking pauses from finished answers. This approach also enables a live filler-word counter from the transcript.
Meridian48 take
A practical reminder that cloud-based speech-to-text is not a real-time sensor; for latency-sensitive signals like silence, raw audio analysis is more reliable.
Read the full reporting
Why I measure interview silence with WebAudio, not the Speech API →
DEV Community
webaudiospeech-recognition