Linus Torvalds 8690477741 Validation: measure [COMPRESSOR] against somebody actually playing ...

A compressor cannot be measured with a steady tone.  It can, and the
answer is exact and worthless: the knee lands where Level says and the
slope above it is 4.81 against the 4.8 on the pot, which tells you the
arithmetic is right and nothing about whether the effect is any use.
What a compressor does is change its gain over time in response to a
signal that changes over time, and a tone that never changes measures
the one case that never happens.

So analyse-compressor.py does both.  The static half is a level sweep
with tones.  The dynamic half plays Inputs/BassForLinus.mp3 - a friend
of Linus's playing scales on a bass, already public elsewhere and
committed here unmodified - and asks what the gain actually did across
seventy-seven seconds.

The decode is a measurement choice rather than a detail, so it is in the
script with its reasons.  The mp3 reconstructs to +2.86 dBFS, because
the master was near full scale and mp3 overshoots, so used raw it clips
the pedal's input before the compressor sees any of it.  It is scaled to
-6 dBFS peak, which is where an instrument track is usually aimed.  A
quieter input crosses the threshold less often and every number moves,
which is why the page says what was chosen.

Only bins with at least two windows are plotted.  An empty bin reported
as zero is not a measurement and would draw a line down to it, and a bin
holding one window is noise wearing a number; the counts are printed so
the thin ends can be read sceptically.

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

2026-08-10 17:38:01 -07:00

..

BassForLinus.mp3

Validation: measure [COMPRESSOR] against somebody actually playing

2026-08-10 17:38:01 -07:00