SmartTurn Turn Node
The Turn node detects whether a speaker's turn is complete in conversational audio. It buffers up to 8 seconds of audio and, when the infer action is called, runs the Smart-Turn ONNX model to produce a turn-completion probability.
| Number of inputs | 1 |
| Number of outputs | 0 |
Configuration​
threshold (Optional)​
Type: Float
The probability threshold above which a turn is considered complete. Default value is 0.5.
modelPath (Optional)​
Type: String
Path to a custom ONNX model file on disk. When not set, the extension uses the embedded model.
Properties​
threshold​
Type: Float
The probability threshold above which a turn is considered complete. Default value is 0.5.
modelPath​
Type: String
Path to the currently loaded model file. Empty string when using the embedded model.
windowDuration (Read-only)​
Type: Float
The maximum audio window duration in seconds. Always 8.0.
Actions​
infer​
Runs the Smart-Turn model on the current audio buffer and emits a turnDetected event with the result.
Parameters​
None
reset​
Clears the internal audio buffer.
Parameters​
None
Events​
turnDetected​
Emitted after each infer action with the turn-detection result.
Data​
| Name | Type | Description |
|---|---|---|
probability | Float | Sigmoid probability that the speaker's turn is complete (0.0–1.0). |
isTurnComplete | Bool | true when probability exceeds the configured threshold. |