level op• Data kinds: signal → table
• Call: import acoustics; acoustics.percentile_level(x, rate, percentiles=(10.0, 50.0, 90.0), weighting='A', ref=1.0, window_s=0.125, floor_db=-200.0) (or opsacoustics.get("percentile_level"))
Statistical levels: `L_N is the level exceeded N` % of the time.
The record is cut into non-overlapping blocks of `window_s` seconds, each
block's equivalent level is computed, and `L_N is the (100-N)`-th
percentile of those levels. Non-overlapping rectangular blocks are used
rather than an exponential time weighting because the block length is then
exactly what the caller asked for and the statistic is exactly a percentile
of the returned `levels` array — an exponential average would make the
effective averaging time a function of the signal.
Returns a dict with one key per requested percentile (`"L10", "L50"`,
`"L90", formatted with %g), plus levels` (the per-block levels),
`times (block start times, s), n_blocks, block_samples`,
`leq (the energy-equivalent level of the whole record), ref`,
`weighting`.
Note that `L50 is the *median* level and leq` is the *energy* level;
they are different numbers whenever the signal is not stationary, and the
gap between them is itself the usual measure of how fluctuating a record is.
Measured on a two-level test signal (1 s at 16 kHz, first half a 1 kHz sine
of amplitude 1.0, second half the same at 0.1, Z-weighted, 0.125 s blocks,
8 blocks): `L10 = -3.010300 and L90 = -23.010300` dB — exactly the two
constituent levels, 20.000000 dB apart, as they must be for a 50/50
split. `L50 = -13.010300` is the interpolated midpoint of the two clusters
and `leq = -5.977386, which is 17 dB above L90`: the energy level sits
near the loud half while the median sits between them. On a
constant-amplitude signal all three percentiles and `leq` agree to
3.6e-15 dB.
Raises `ValueError: everything :func:_as_signal` refuses, a
percentile outside `[0, 100], a non-positive window_s, a window_s`
longer than the record (which would give one block and make every percentile
the same number while still looking like a statistic), `ref <= 0`.
• acoustic_condition_monitoring family guide
• Sample-data catalog (download URLs / licences) — 2-D uses skimage.data (BSD/public domain) plus synthetic images; 3-D lists download URLs for real data sources (Stanford, PDS, …).
• Operator provenance and references — the sources of the research/methods this op family came from.
• The canonical algorithm (author, year) and its uses are named in the family usage guide above.
• acoustic_condition_monitoring — py -3.11 examples/acoustic_condition_monitoring.py
table as input)level)octave_bands · octave_spectrum · weighting_response · apply_weighting · equivalent_level
*Provenance: acoustics.py — ACOUSTICS operator registry. This per-op note is generated by tools/opdocs.py md (do not hand-edit).*
© 2026 Kazufumi Furuse — Fullseye operator documentation. Licensed under Apache-2.0.