Skip to main content

SmartTurn Extension

The SmartTurn extension detects the end of a speaker's turn in conversational audio. It uses an on-device ONNX model to compute a probability that the current speaker has finished talking, enabling conversational AI applications to know when to respond.

The model accepts up to 8 seconds of 16 kHz mono audio and produces a turn-completion probability score. It works well in combination with a voice activity detector (e.g. SileroVAD) — use VAD to detect when speech ends, then call infer to decide whether the speaker's turn is truly complete.

The underlying model is based on the smart-turn project by pipecat-ai.

Dependencies​

The SmartTurn extension requires the ONNX extension to be loaded.

Nodes​

The SmartTurn Extension provides the following audio nodes for a Switchboard SDK audio graph:

NodeDescription
TurnDetects end-of-turn in conversational speech.

Download​

You can find the download links for this extension on our Downloads page.