Skip to main content

noise_floor_min_stat

Function noise_floor_min_stat 

Source
pub fn noise_floor_min_stat(energies: &[f32], window: usize) -> f32
Expand description

Estimate the noise floor as the bias-compensated minimum energy over the trailing window entries of energies (per-frame RMS or power values).

  • energies: recent per-frame energy statistics, oldest→newest. Only the last window entries are considered; if fewer are present, all are used.
  • window: number of trailing frames the minimum is taken over. A larger window is more robust to sustained speech but slower to track rising noise.

Returns the estimated floor min(recent) * NOISE_BIAS, clamped to be non-negative. Returns 0.0 for an empty slice (no evidence yet — the caller should treat a zero floor as “unknown”, which downstream gates guard with a tiny epsilon rather than declaring everything voiced).

Non-finite inputs (NaN/±inf) are ignored so a single bad sample cannot poison the estimate.