Struct TurnController 

pub struct TurnController<T: AudioTurnDetector> { /* private fields */ }

Orchestration wrapper around any AudioTurnDetector.

Tracks prediction state across calls and provides convenience methods like reset_if_finished for correct VAD integration without manual state bookkeeping.

Usage

let detector = PipecatSmartTurn::new()?;
let mut ctrl = TurnController::new(detector);

// Audio arrives continuously
ctrl.push_audio(&frame);

// VAD speech start — soft reset (keeps buffer if turn was unfinished)
ctrl.reset_if_finished();

// VAD speech end — predict
let result = ctrl.predict()?;

See reset_if_finished for details on when to use soft vs hard reset.

Implementations

impl<T: AudioTurnDetector> TurnController<T>

pub fn new(inner: T) -> Self

Create a new controller wrapping the given detector.

pub fn push_audio(&mut self, frame: &AudioFrame<'_>)

Feed audio into the detector.

pub fn predict(&mut self) -> Result<TurnPrediction, TurnError>

Run prediction on buffered audio.

Tracks the result state internally for reset_if_finished.

pub fn reset(&mut self)

Hard reset — always clears the buffer.

Use when you know a new turn is starting (e.g. after the assistant finishes responding).

pub fn reset_if_finished(&mut self) -> bool

Soft reset — clears the buffer only if the last prediction was Finished or no prediction has been made since the last reset.

Returns true if a reset occurred, false if skipped.

Call this on VAD speech-start when you don’t know whether the user is continuing the same turn or starting a new one. If the previous prediction was Unfinished, the buffer is preserved so the next predict runs on the full accumulated audio.

pub fn last_state(&self) -> Option<TurnState>

Returns the state from the last predict call, or None if no prediction has been made since the last reset.

pub fn inner_mut(&mut self) -> &mut T

Returns a mutable reference to the inner detector.

pub fn into_inner(self) -> T

Unwrap the controller, returning the inner detector.