Struct StreamWindow 

pub struct StreamWindow {
    pub samples: Vec<f32>,
    pub real_samples: usize,
    pub window_start_secs: f32,
    pub had_speech: bool,
    pub trailing_silence_secs: f32,
    pub cap_hit: bool,
    pub forced_eou: bool,
}

Fields

samples: Vec<f32>

Exactly max_window_secs × 16 000 samples; zero-padded at the BACK when the growing buffer hasn’t reached max_window_secs yet.

real_samples: usize

Number of real (non-padding) samples in samples.

window_start_secs: f32

Wall-clock offset since stream start of the first sample in the current utterance.

had_speech: bool

True if any VAD-positive frame falls within the current utterance buffer.

trailing_silence_secs: f32

Seconds since the last VAD-positive frame; for use by the endpointer.

cap_hit: bool

True when the buffer hit max_window_secs. The consumer must call either Chunker::trim_oldest or Chunker::reset_utterance before the next push; otherwise the next cap will additionally set forced_eou=true.

forced_eou: bool

True only on the second consecutive cap without an intervening trim or reset. Treat as a forced EOU (finalize + reset_utterance) — the safety-net path when the trim wasn’t applied (e.g. no prior commits to anchor against).