Level

Reporting item 1 — LAeq + dynamic range. Plus crest factor as a diagnostic.

from debussy.level import laeq_dbfs, dynamic_range_db, a_weighting_filter

laeq_dbfs(y, fs) -> float

A-weighted equivalent continuous level, in dBFS-A — full-scale relative, not absolute SPL. The A-weighting filter follows IEC 61672-1, applied as a second-order-section cascade.

Digital audio carries no absolute reference, so an uncalibrated LAeq is only comparable within a set of files produced the same way. For true dB SPL, measure a reference tone through your playback chain and pass the difference as calibration_offset_db to analyze_audio(); it is added to LAeq and recorded in Result.notes.

dynamic_range_db(y, fs, win_ms=50.0) -> float

The 95th minus the 5th percentile of short-term RMS level, on 50 ms windows with 50 % overlap. Percentiles rather than peak-to-trough, so a single click does not dominate. Returns 0.0 for signals shorter than one window.

Lower values mean a more even stimulus; the companion review treats a stable mean level with minimal transients as part of the Tier-1 event-structure check.

a_weighting_filter(fs)

Returns the SOS coefficients used above, if you want to apply the weighting yourself.

Crest factor

Not a separate function — analyze_audio() computes it as 20·log₁₀(peak / RMS) and reports it in Result.crest_factor_db. It is a diagnostic rather than a guideline item: high crest with low dynamic range indicates isolated transients on an otherwise even bed.

Long files

For recordings past the full-analysis threshold, level metrics are not estimated from probes — they are computed exactly over the whole file by streaming it in blocks. A long ambient piece reports the same LAeq and dynamic range it would if analysed in one pass.