Psychoacoustic

Reporting item 3 — roughness and item 6 — sharpness. Both delegate to MOSQITO for the validated reference implementations.

from debussy.psychoacoustic import psychoacoustics

psychoacoustics(y, fs, suppress_warnings=False) -> dict

One call returns both metrics:

Key	Metric	Model
roughness_asper	Roughness, asper	Daniel & Weber (1997), mosqito.roughness_dw
sharpness_acum	Sharpness, acum	DIN 45692, mosqito.sharpness_din_st
roughness_coverage_pct	% of duration above 0.3 asper	derived from the roughness time series

Either value may be None on inputs where the model degenerates — a pure tone has no amplitude modulation for the roughness integrator to work on. The failure reason is recorded in Result.notes rather than raised.

Roughness is the Tier-1 gate

Roughness scored highest of all twelve reviewed parameters (12/12) and is the strongest single Tier-1 constraint: 30–150 Hz amplitude modulation drives amygdala threat-detection responses. The target is mean < 0.3 asper.

roughness_coverage_pct exists because a mean can hide things. A stimulus that is calm for four minutes and harsh for ten seconds can average below threshold while still containing the passage that matters. The Tier-1 verdict therefore takes the more severe of the whole-file mean and the proportion of time above 0.3 asper — coverage can only make a verdict more conservative, never rescue a high mean.

The 2 % / 10 % coverage bands are provisional screening heuristics, not validated cut-offs.

Sample rate

MOSQITO's models require 48 kHz and resample internally when given anything else. DEBUSSY emits a UserWarning when that happens — suppress_warnings=True silences it. Result.sample_rate always reports the file's native rate.

Dependency note

MOSQITO imports matplotlib inside both of these calls without declaring it, so matplotlib is a required dependency of DEBUSSY. Without it both metrics silently returned None before v0.2.1.