transform op• Data kinds: none → table (an op determined by its arguments alone — it takes no image or data input)
• Call: import acoustics; acoustics.stft_cola_check(window='hann', win=256, hop=None) (or opsacoustics.get("stft_cola_check"))
Does this (window, hop) pair satisfy COLA, and how exactly?
COLA — the analysis windows summing to a constant over the hop lattice — is
what lets plain overlap-add work without a division. It is *not* required by
:func:istft, which is weighted, but it is required by anything that
overlap-adds modified frames without renormalising, and getting it wrong
produces a periodic amplitude ripple at `rate/hop` Hz that looks like
tremolo rather than like a bug.
Returns a dict: `cola (bool), constant` (the mean of the overlap sum),
`max_deviation (absolute), relative_deviation, nola` (bool),
`min_squared_sum`, plus the geometry.
Measured (periodic windows, `relative_deviation` of the plain sum):
======== ==== ==== ================== ======== ====
window win hop relative_deviation constant COLA
======== ==== ==== ================== ======== ====
hann 256 128 4.44e-16 1.00 yes
hann 256 64 2.22e-16 2.00 yes
hann 256 85 1.48e-03 1.506 no
hamming 256 128 2.06e-16 1.08 yes
blackman 256 128 1.91e-01 0.84 no
blackman 256 64 3.97e-16 1.68 yes
boxcar 256 128 0.00e+00 2.00 yes
======== ==== ==== ================== ======== ====
The two blackman rows are the useful ones: the same window is COLA at
hop = win/4 and 19 % off at hop = win/2, so "which window" is not the
question — the pair is. The boxcar row was worth measuring rather than
assuming: a rectangular window at 50 % overlap sums to exactly 2 and is
COLA, which is the opposite of the usual intuition about it. Note also that
the constant is not 1 in general — an overlap-add that does not divide by it
is off by a *gain*, which is the failure that looks like a working system.
Raises `ValueError: unknown / all-zero window, hop` outside
`[1, win], win outside [2, MAX_WINDOW]`.
• 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)transform)*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.