Expand description
Segmental signal-to-noise ratio (seg-SNR) between a reference and a degraded signal.
Classic intrusive metric: split both signals into fixed-length frames, compute the per-frame SNR in dB (reference energy over error energy), clamp each frame to a sensible dynamic range, and average across frames. Unlike a global SNR, the segmental variant weights quiet and loud passages more evenly, which correlates better with perceived quality.
Pure-Rust, no licence encumbrance, zero heap: a single scalar f32 is
returned and the two input slices are read in place.
Constants§
- MAX_
FRAME_ SNR_ DB - Per-frame SNR is clamped to
[MIN_FRAME_SNR_DB, MAX_FRAME_SNR_DB]before averaging. These match the conventional seg-SNR range used in speech-quality literature and keep silent/identical frames from dominating the mean with+inf. - MIN_
FRAME_ SNR_ DB
Functions§
- segmental_
snr - Segmental SNR in dB between
referenceanddegraded.