What Is Voice Activity Detection?

Voice Activity Detection Explainer

Voice activity detection, or VAD, classifies short regions of audio as speech or non-speech. It produces a timeline that indicates when a person is likely speaking, without transcribing the words or determining the speaker’s identity.

Simple detectors use energy, spectral shape, or periodicity. Learned systems use labeled recordings to distinguish voice from background sounds and may use surrounding frames to avoid rapidly switching on and off.

VAD reduces unnecessary transmission in calls, trims silence before speech recognition, helps segment recordings, and serves as an early step in diarization, enhancement, and voice interfaces. Real-time use requires quick decisions and low computation.

Quiet speech, laughter, singing, breath sounds, music, television, and changing noise can confuse a detector. Thresholds trade missed speech against false alarms. Testing should cover languages, ages, microphones, distances, environments, and the latency acceptable for the application.

Voice activity detection estimates which moments in an audio stream contain speech and which contain silence or other sounds.

Key terms include voice activity detection, speech frame, non-speech, energy threshold, hangover, false alarm, missed speech and latency.

The detector marks short time frames that likely contain a human voice.

Smoothing across nearby frames prevents brief pauses from fragmenting one utterance.

A sensitive setting captures quiet speech but may trigger on background sound.

A strict setting reduces false alarms but can cut off soft words or sentence endings.

Explore more "Explainers"

Discover additional explainers across politics, science, business, technology, and other fields. Each explainer breaks down a complex idea into clear, everyday language—helping you better understand how major concepts, systems, and debates shape the world around us.