fn window_bounds(pcm: &[f32]) -> Vec<(usize, usize)>Expand description
Sample ranges tiling pcm, each at most WINDOW_TARGET long,
cutting at the quietest ENERGY_FRAME block between
WINDOW_MIN and WINDOW_TARGET past the previous cut. A take
that fits one window comes back whole, which is the exact buffer
the recognition runtime saw before windowing existed.
Audio with no dip anywhere still splits, at the ceiling: the windows tile exactly either way, so a word straddling such a cut may transcribe loosely across the join but no audio is dropped.