Interface OpenAIRealtimeTurnDetection

Server turn detection for a transcription session: OpenAI's voice activity detector decides where each utterance ends and commits it.

interface OpenAIRealtimeTurnDetection {
    threshold?: number;
    prefixPaddingMs?: number;
    silenceDurationMs?: number;
}

Properties

threshold?: number

Activation threshold in [0, 1]; a higher value needs louder speech.

Default Value

0.5
prefixPaddingMs?: number

Audio kept before detected speech, in milliseconds.

Default Value

600
silenceDurationMs?: number

Silence that ends an utterance, in milliseconds.

Default Value

350