Struct AudioRecorder 

pub struct AudioRecorder {
    pub sample_rate: u32,
    /* private fields */
}

Accumulates inbound (caller) and outbound (bot) audio on a shared wall-clock timeline and mixes them down to a single mono WAV byte buffer.

Each leg records only its active chunks, but every chunk is stamped with its arrival time relative to call start, so render silence-pads the gaps and the two legs stay aligned — a caller turn at t=2s and a bot turn at t=5s land where they actually happened instead of both starting at sample 0 (which made the two voices play over each other in the recording).

Fields

sample_rate: u32

Output WAV sample rate in Hz.

Implementations

impl AudioRecorder

pub fn new(sample_rate: u32) -> Self

Create a recorder that renders mixed audio at sample_rate Hz.

pub fn push_inbound(&mut self, chunk: &AudioChunk)

Append a chunk from the inbound (caller) leg, stamped at the live wall clock.

Stored at its own rate; resample + silence-pad happen in render_wav. Empty chunks are dropped.

pub fn push_outbound(&mut self, chunk: &AudioChunk)

Append a chunk from the outbound (bot) leg, stamped at the live wall clock.

pub fn push_inbound_at(&mut self, offset: Duration, chunk: &AudioChunk)

Append an inbound chunk at an explicit timeline offset from call start.

Deterministic entry point (no wall clock): the live taps go through push_inbound; tests place chunks at known offsets.

pub fn push_outbound_at(&mut self, offset: Duration, chunk: &AudioChunk)

Append an outbound chunk at an explicit timeline offset from call start.

pub fn duration_seconds(&self) -> f64

Wall-clock call duration in seconds — the end of the later leg’s last audio. Reported back to the control plane as usage_metrics.duration_seconds.

pub fn render_wav(&self) -> Result<Vec<u8>, FlowcatError>

Mix both legs to mono and render a complete WAV file as bytes.

Each leg is resampled to self.sample_rate and laid out on the shared timeline (silence-padding the gaps between turns); the two streams are summed sample-by-sample (zero-padding the shorter leg to the longer), each sum is clamped into i16, and the result is written as a mono 16-bit PCM WAV via hound.