Glossary
VAD (voice activity detection)
Identifying which portions of an audio signal contain speech and which are silence, music, or noise — usually the first stage of a speech-processing pipeline.
VAD marks speech regions so downstream stages — ASR, diarization, speaker identification — process only what matters. Its errors compound: clipped word onsets raise WER, and missed speech inflates diarization error. It also drives duration accounting: hours of speech in a dataset spec usually means detected speech time, not raw file length.
Why it matters
Speech-hour counts rest on VAD, so it quietly defines the quantity a buyer is paying for.