Skip to content

OpenAIRealtimeSemanticVad

Semantic VAD turn-detection configuration for OpenAI Realtime transcription sessions.

Defined in: src/providers/stt/openai/OpenAIRealtimeSTT.ts:57

Semantic VAD turn-detection configuration for OpenAI Realtime transcription sessions.

Remarks

Semantic VAD uses a turn-detection model (in conjunction with VAD) to estimate whether the user has finished speaking, waiting longer when the utterance sounds incomplete (e.g. trails off with “uhhm”).

See

OpenAIRealtimeSTTConfig.turnDetection

Properties

PropertyTypeDescriptionDefined in
eagerness?"auto" | "low" | "medium" | "high"How eagerly to end the turn. 'low' waits longer for the user to continue; 'high' chunks audio as soon as possible. Default 'auto' (equivalent to 'medium')src/providers/stt/openai/OpenAIRealtimeSTT.ts:65
type"semantic_vad"Model-based semantic turn detection.src/providers/stt/openai/OpenAIRealtimeSTT.ts:59

© 2026 CompositeVoice. All rights reserved.

Font size
Contrast
Motion
Transparency