The short answer
Sidechain compression means driving a compressor from a separate detection signal rather than reacting only to the audio it is turning down. A compressor decides how hard to clamp by listening to a signal and measuring how far it crosses the threshold. Sidechaining is nothing more than changing what that detector listens to. The audio you hear does not have to be the same thing the compressor reacts to.
There are two different things people mean by the term, and they solve different problems. One reshapes what the compressor reacts to inside a single channel, so it stops pumping on bass. The other feeds the detector from a completely different source, so one sound ducks another. This page assumes you already know what a compressor does; if you need the basics, start there and come back. Here we go one level down, into the detection signal every compressor is quietly built around.
What a compressor actually listens to
Inside any compressor two jobs happen at once. One path is the audio you hear, passing through a gain stage. The other is a detector: a meter that watches a signal, measures how far it goes over the threshold, and tells the gain stage how much to turn down. Normally the detector watches a copy of the same audio that is passing through, so the compressor reacts to the very thing it is compressing. That copy is the detection signal, also called the sidechain or the key.
The word sidechain is just the name for that side path. It is called side because it sits beside the main audio path and never reaches your ears directly. It only steers the gain. Once you see the detector and the audio as two separate signals, the whole idea falls out on its own: you can change the detection signal without changing the audio at all. What the compressor reacts to and what you hear are allowed to be different things.
Meaning one: filtering the detection signal
The most common and most useful form of sidechaining stays inside one channel. You take the detection signal and run it through an EQ-style filter, usually a high-pass, before the detector sees it. The audio you hear is untouched and still full-range. Only the copy feeding the detector loses its low end.
Why bother? Bass carries most of the raw energy in a mix. A kick or a low note is far louder, in level, than the words riding on top of it. A compressor watching the full-range signal sees those bass hits as the loudest thing around and clamps the entire signal every time one lands. The result is audible pumping: the whole mix ducks and swells in time with the low end. High-pass the detection signal and the detector stops seeing the bass. It reacts to the midrange and the voice instead, so it controls what you actually wanted to control and leaves the lows to breathe.
A low-pass on the detection signal does the mirror-image job. It stops very high frequencies, like harsh s sounds, from over-triggering the compressor, so it does not clamp down and emphasize sibilance. Used together, a high-pass and a low-pass form a band-pass on the detector, letting you point the compressor at exactly the slice of the signal you want it to respond to. Through all of it, the audio itself stays full-range. You are only ever filtering the trigger.
Meaning two: ducking one sound under another
The second meaning is the famous one: feed the detector from a completely different source than the audio being compressed. This is called external-key sidechaining, or just ducking. The classic example is a voiceover. Put a compressor on the music, but instead of the music driving its own detector, feed it the microphone. Now the music turns itself down whenever the voice speaks and comes back up in the gaps. The music never triggers itself; the voice triggers it.
The dance-music pumping effect is the same trick with a kick drum as the key: every kick clamps the synths, so the whole track breathes in time. Ducking is powerful, but notice what it needs: a connection from one channel into another channel's compressor. It is as much a routing move as a compression move, which is why it lives in the world of audio routing and not just the compressor panel.
Why high-passing the detector stops pumping
A worked example makes it concrete. Say your voice sits around -18 dB and your threshold is set there, but a low rumble or a plosive occasionally spikes to -6 dB. To the detector that spike is 12 dB over threshold, so it clamps hard, dragging your whole voice down with it for as long as the release takes to recover. Because the spike is brief and the release is comparatively slow, you hear the level dip and swell back up. That is pumping.
Raise the high-pass on the detection signal until the detector no longer sees that low rumble, and the spike stops registering. The threshold now applies to the part of the signal you care about. The audio still carries its full low end to the listener; you only changed what the compressor was reacting to. This is the honest fix for a compressor that will not stop breathing on bass, and it is far more surgical than simply raising the threshold, which would just let more of everything through.
The detector and the gain, kept separate
The whole mechanism fits in one line: a compressor computes its gain from the detection signal, then applies that gain to the audio. Those are two different signals wearing the same name only by default.
out = audio * gain( level( detection ) )The gain reduction is a function of the DETECTION signal's level. Filter the detection and you change the gain without ever filtering the audio. Feed the detection from another channel and that channel drives this one.
That one separation, detection versus audio, is the whole of sidechaining. Everything else is a choice about what to put in the detection slot: the same signal for ordinary compression, a filtered copy of it for sidechain filtering, or a different signal entirely for ducking. Change the input to level() and you change how the compressor behaves without ever touching the sound that passes through.
Sidechain in patchd
In patchd the compressor is a free node you add to any channel, and it carries the filtering kind of sidechain built in. Two controls shape the detection signal without ever touching your audio:
- Sidechain HP (20 Hz to 2000 Hz, default 20 Hz). A high-pass on the detection signal only. Left at 20 Hz the detector hears everything; raise it toward a few hundred Hz and low-frequency energy stops triggering gain reduction, so the compressor stops pumping on bass and plosives.
- Sidechain LP (1000 Hz to 20000 Hz, default 20000 Hz). A low-pass on the detection signal only. Left at 20 kHz nothing is removed; lower it and high-frequency energy, like sibilance, stops over-triggering the compressor.
Both filters act purely on what the compressor listens to. Your audio always passes through full-range; these controls never EQ the sound itself, only the trigger. Together they are a band-pass on the detector, so you can aim the compressor at exactly the part of the signal you want it to respond to and leave the rest alone.
One thing to be clear about: patchd's compressor sidechain is the filtering kind, inside a single channel. Triggering one channel's compressor from a different channel, the cross-channel ducking described above, is a routing setup patchd's compressor does not do. If you came here to duck music under a voice automatically, that specific feature is not part of the compressor today.
Sidechain filtering pairs naturally with sensible attack and release settings: filter the detector so it only sees what matters, then set the timing so it responds smoothly. patchd is in development now, with the compressor and the rest of the dynamics set free on every channel. Join the waitlist and we will tell you the moment it is ready to install.