Voice activity detection (VAD)

Voice activity detection is the lightweight signal-processing step that determines whether incoming audio contains speech — used to start STT, trigger barge-in, and gate microphone use in voice agents.

VAD runs continuously on the user's microphone audio, classifying each ~20ms frame as speech or non-speech. Used to start STT when the user begins talking, end STT when they stop (with a configurable silence threshold), trigger barge-in mid-response, and avoid sending silent audio to expensive STT APIs. Modern VAD models (Silero VAD, WebRTC VAD, Picovoice Cobra) are extremely lightweight — single-digit milliseconds per frame, runnable on-device. Best practice: pair VAD with an STT model's own end-of-utterance detection for robust turn-taking.

When to use voice activity detection (vad)

• Any realtime voice agent.
• Battery-sensitive on-device voice apps.

Common mistakes

• Tuning VAD too aggressive — clips the start of user speech.
• Treating VAD as a hard signal — pair with model-level end-of-utterance detection.

FAQ

Related terms

• Voice (LLM apps) — Voice in LLM apps refers to the full speech pipeline — speech-to-text (STT), language model, text-to-speech (TTS) — that lets users converse with an AI assistant in spoken language.
• Streaming STT — Streaming STT (speech-to-text) emits partial transcriptions as the user speaks — instead of waiting for end-of-utterance — enabling sub-second response from a voice assistant.
• Barge-in — Barge-in is the voice-agent feature where the user can interrupt the assistant mid-response — the assistant detects the speech and stops talking — making conversations feel natural instead of robotic turn-taking.
• AI agent — An AI agent is a system where a language model autonomously plans and executes a sequence of tool calls to accomplish a goal.

Last updated: 2026-06-01. Raw markdown: https://promtable.com/glossary/vad.md.