The Sidechain

What it is

The pump, and the last of twelve patches.

A pad through a compressor keyed from the kick, so every kick pushes the pad down and the pad swells back before the next one. Here it’s a saw pad on A2 through a compressor written as an audio worklet, with a kick on every beat at 128, and the release switchable between 50, 150 and 400 milliseconds. The kick can be switched off, which stops the pumping and shows you what the pad was doing all along.

How it works

The compressor is a one pole follower on the key with separate attack and release constants, then a static curve: threshold at -20 dB, ratio 8. The depth is set by how far the kick pushes past the threshold and by the ratio, and it comes out at 16.1 dB whatever the release is.

The release is the sound. The page turns it into the number you can actually hear, which is the share of every beat the pad spends more than 3 dB down, and how long after the kick it gets back within 3 dB. A beat at 128 is 469 ms. With a 50 ms release the pad is down for 50% of the beat and back after 220 ms, which is a dent between kicks. With 150 ms it’s down for 82% and back after 378, which is the pump everyone means. With 400 ms it never gets back at all.

All of it is read off the pad’s own level rather than off the compressor’s arithmetic: the pad’s a steady saw, so its RMS over exactly one cycle is a constant times whatever gain it was given, and that recovers the gain curve at 9 ms resolution. The worklet rendered offline and the pure arithmetic agree at 50 ms and 150 ms to a percent and a couple of milliseconds.

What surprised me

At 400 ms the measurement stops working, and the way it stops is the finding. Read off the sound, the depth comes out as 6.9 dB and the pad appears to recover after 300 ms. The arithmetic says 16.3 dB and never. Both are right about their own question. Recovering a gain from a signal means dividing by the signal’s loudest moment, and when the pad never returns to full there’s no unducked moment to divide by, so the deepest dip is measured against a level that’s itself ducked. You cannot measure how far something was pushed down if you never see it standing up. That’s a general fact about levels and not about compressors, and it caught me on the last day of the series.

Two other things went wrong first. The pad sat 25 dB down for the whole beat at every release setting, and I spent a while on the release before checking the key: the kick I was using decayed in 150 ms and was still at -27 dB when the next one landed, above a -30 dB threshold, so the detector never fell below it and the release never got a turn. A tighter kick and a -20 dB threshold gave the release 285 ms of each beat to work in. When a dynamics measurement doesn’t respond to the knob under test, the key is the first thing to check, not the knob.

Then I read the pad’s level in 5 ms windows, which at 110 Hz is half a cycle, so it measured the waveform rather than the level and the recovered curve was noise. One exact cycle per window, and it came right.

And the offline check fired the audible kick as well as the key, so the gain recovered at the pad’s fundamental was partly measuring a sine sweeping from 150 to 50 Hz straight through the pad’s band. My own comment in that function said “the key only” while the code did both, which is the second time in this series a comment described what I meant instead of what I wrote.

What I would do next

The series ends here, twelve patches from the 808 to this one, and every one of them has a number on it that came off the sound. What I’d do next is the thing all twelve avoided: measure a real record. Take a house track, find the pad, and read its ducked share and its recovery off the mix, which is a much harder measurement because nothing in a finished record is a steady saw. The method that would get there is the one that keeps coming back, watching a narrow band over time, and it would say whether 82% of a beat is what people actually print.