What is parallel compression?

Parallel compression in plain English

Normal compression puts a compressor in the path of a sound, so everything you hear has been through it. Parallel compression splits the sound in two. One path stays exactly as it was, and the other goes through a compressor set far harder than you'd ever use on its own. Then the two get mixed back together, with the squashed copy usually sitting some way under the original.

Each path covers for the other. A transient, the short spike at the very start of a drum hit or a plucked note, comes through the dry path at full height, because nothing there touched it. The copy has had its transients flattened, but its quiet parts have been turned up a long way, like the ghost notes on the snare and the room sound after each hit. Add the two and you keep the attack of the original with the density of the copy underneath.

It's also known as New York compression, a name tied to the New York studios where engineers made it a regular trick on drums. The technique is the same whatever you call it, and every DAW can do it.

If compression itself is new to you, start with what is compression?, which covers threshold, ratio, attack and release from the beginning. This page assumes you know roughly what those controls do and goes into what changes when you run a compressor alongside the dry sound instead of in its way.

Parallel compression in numbers

A normal compressor works from the top down: it turns the loud parts down and leaves the quiet parts alone. Parallel compression ends up working from the bottom up. That sounds odd until you follow the levels through.

Take the settings the lab uses: a −30 dB threshold, a 10:1 ratio and 20 dB of makeup gain on the copy. The quiet room sound after a hit, at −50 dB, is under the threshold, so the compressor leaves it alone and the makeup gain lifts it to −30 dB. A snare peak at −6 dB is 24 dB over the threshold, so at 10:1 it comes out 2.4 dB over, at −27.6 dB, and the makeup brings it to −7.6 dB. In the copy, the room sound and the snare are now 22 dB apart, where in the original they were 44 dB apart.

Now blend the copy in at 30%. At the snare peak, the copy comes out only 1.6 dB under the original, so mixing the two hardly changes the level: the peak ends up at −6.4 dB, 0.4 dB lower than it started. The room sound is a different story, because the copy has it 20 dB louder than the original does, and in the blend it comes out at −38.6 dB, 11.4 dB higher than before. The gap has shrunk from 44 dB to about 32 dB, and nearly all of that came from turning the quiet part up.

With these settings there's a crossover at about −8 dB, where the copy comes out at exactly the level of the original. Anything quieter than that gets lifted by the blend, and the further under the crossover it is, the more it gets lifted. Peaks near the crossover are left almost exactly where they were. Drag the mix slider in the figure and you can see the bottom of the curve rise while the top stays near the diagonal.

That bottom-up shape is the reason to do it. The loudest moments, which are the parts most easily damaged by heavy compression, are left almost alone, while the quiet parts, which are the ones that get lost in a busy mix, come up. Getting the same lift from a single compressor in series would mean squashing the transients too.

Reading the lab

The lab is the same compressor as the one on the compression page, with the knee locked at 0 dB so the numbers line up with the figure above. The lesson starts with the drums dry, builds a crushed copy, then blends it back under the original with the Mix slider.

On the Whole loop picture, the pale shape is the original loop and the pink shape is what comes out, which below 100% mix is the blend. The amber gain reduction lane shows what the compressor is doing to the copy, so expect it to dip a long way, 20 dB or more on the loud hits. That number would be alarming on a compressor in series; on a parallel copy it's normal. The thin spikes still poking out at the start of each hit in the copy are the fraction of a millisecond before the compressor catches up, which only a compressor with lookahead can remove.

The Transfer curve picture draws the compressed copy in pink, with a dashed line for the blend below 100% mix. The Original and Compressed switch takes you back to the dry loop at matched loudness, so you can hear what the blend changed without the extra level fooling you.

Mix knob or a separate bus

There are two ways to do this in a DAW, and they sound the same once the levels are set.

With a mix knob

Many modern compressors have a control that blends the compressed and dry signals inside the plug-in. Ableton Live's Compressor calls it Dry/Wet and Logic's Compressor calls it Mix. Turn the compressor up hard, then bring the mix down until the original comes back on top. Because both paths live inside one plug-in, they can never drift out of time.

A mix knob crossfades, though. At 30% you hear 30% of the copy and 70% of the dry, so the dry path itself is turned down a little as you raise the mix. The lab's Mix slider works this way, and so does the figure above.

With a send to a separate bus

The older way, and the one you'll need with a compressor that has no mix knob, is a send to a separate bus. The track plays as normal, a send copies it to a return or aux channel, and the compressor sits on that channel at 100% wet. The dry track stays at full level and the bus fader decides how much of the copy you add.

The bus takes a bit longer to set up and pays you back in flexibility. Several tracks can feed one parallel bus, which is how a drum kit usually gets it, with kick, snare, toms and room mics all going to the same squashed copy. You can also EQ or saturate the copy without touching the original.

To move between the two, you need to know what a mix setting means as a bus level. A 30% mix is the same blend as a bus sitting 7.4 dB under the dry track, with the whole result 3.1 dB quieter. Run the bus method at that −7.4 dB level and the snare peak from earlier rises by 2.6 dB while the room sound rises by 14.4 dB. The difference between the two is 11.8 dB either way, so it's the same shape at a different overall level.

These figures assume a straight linear crossfade, which is what the lab uses. Some plug-ins use a different curve, so treat a mix percentage as a starting point and set the final blend by ear.

How to set up parallel compression

This is the bus method on a drum kit. The same steps work on a vocal or a bass, with gentler settings from the table further down.

1. Make a return or aux busIn Ableton it's a return track, in Logic an aux, and in FL Studio a mixer insert that the drums are routed to. Name it something you'll recognise, like Drums Parallel.
2. Send the drums to itAdd a send from each drum track, or from the drum group, set at 0 dB. A post-fader send keeps the copy following your drum faders, which is usually what you want.
3. Pull the bus fader all the way downStart from silence so you hear the copy arrive, instead of judging a loud blend from the first second.
4. Put a compressor on the bus and set it hardA ratio of 8:1 or more, the fastest attack, a release around 100 ms and the threshold low enough for 10 to 20 dB of gain reduction on the loud hits. Solo the bus while you do this; it should sound flat and dense.
5. Add makeup gain on the busBring the copy up to roughly the loudness of the dry drums, so the bus fader starts from a sensible place.
6. Unsolo and raise the bus faderBring it up slowly under the dry drums. Stop when the quiet hits and the room come forward and the kit feels thicker, then back it off a decibel or two.
7. Check it at matched loudnessMute and unmute the bus while turning the drum group down to make up for the level the copy added. If the drums only sound better when they're louder, pull the bus back.

Settings for the squashed copy

The copy has a different job from a normal compressor, so its settings look extreme.

Threshold and ratio

Set the threshold low, so most of the sound goes over it, and the ratio high, so whatever goes over gets flattened. On drums, 10 to 20 dB of gain reduction on the loud hits is normal. The copy should sound almost wrong when you solo it, because its only job is to bring the quiet parts up close to the loud ones.

Attack

Keep the attack fast, around 0.1 to 1 ms on drums. The dry path is already supplying the transients, so you don't need the copy to let them through. A slow attack on the copy lets the spike of every hit through, boosted by the makeup gain, which piles more level onto the peaks, the one place you didn't want it. On vocals and bass, a few milliseconds is gentler and still does the job.

Release

Release decides how much of each hit's tail comes up. A short release, 30 to 60 ms on drums, lets the copy recover before the next hit, so the tail and the room after every hit get pulled up and the kit sounds busy and aggressive. A long release of 300 ms or more keeps the copy pressed down between hits, which sounds smoother, and around 100 ms is a sensible middle. Watch the release on bass: below about 50 ms the compressor starts to follow the waveform itself and distort.

Blend

Bring the copy up until you can hear it, then pull it back until you can only just hear it. On drums that usually lands with the bus somewhere between 6 and 12 dB under the dry kit. Past a point, the copy starts to set the level of the peaks as well, and then you're back to ordinary heavy compression.

Parallel compression on drums

Drums are where most people first hear it, because the difference between their loud and quiet parts is so large. A snare hit can peak 20 to 30 dB above the ghost notes around it, and the room sound after each hit is quieter still. A compressor in series could close that gap, but only by squashing the hit along with it. In parallel, the hit stays sharp and everything around it comes up.

On a full kit, the usual move is to send the whole drum group, room mics included, to one bus. If the kick drives the bus too hard, high-pass the copy at 80 to 100 Hz, either before the compressor so the kick triggers it less or after it so the copy adds no extra low end. A fast, aggressive compressor suits this bus, and an 1176 with all four ratio buttons pressed in, a mode known as all-buttons, is a common choice, either the hardware or one of the plug-in versions. The drum bus lesson covers what else usually goes on the drum group, and how to make drums hit harder puts parallel compression alongside the other ways of adding punch.

Pick Percussion in the lab to hear the same settings on a looser groove, where the softer notes between the accents come up under them.

Parallel compression on vocals

A lead vocal often needs to sit at one level in a dense chorus while still sounding like a person singing. A single compressor working hard enough to hold it there can make the voice sound pinned in place. A parallel copy fills in the quiet syllables and the ends of words, while the dry vocal keeps its natural shape on top.

Use gentler settings than on drums: a 4:1 to 8:1 ratio, a few milliseconds of attack and 8 to 12 dB of gain reduction, blended low. You can brighten the copy with a high shelf, or add a little saturation to it, so it adds presence as well as level; a bus makes that easy because none of it touches the dry vocal. Pick Vocal in the lab and set the mix around 20% to hear it.

Parallel compression on bass

On bass, a parallel copy evens out notes that ring at different levels while the dry path keeps the pick or the pluck at the front of each one. Keep the release on the copy at 100 ms or longer, because a 55 Hz note takes about 18 ms per cycle and a release near that length can follow the waveform up and down and buzz. A slightly slower attack on the copy, 5 to 10 ms, also stops it flattening the start of each note.

Parallel or serial compression?

Serial compression means two compressors one after the other, each doing part of the work, while parallel compression is one compressor working very hard on a copy. Serial wins when you need firm control of the peaks, like a vocal whose loudest words jump 15 dB above the rest. When the peaks are fine and you want the quiet parts up, parallel is better. Plenty of mixes use both on the same vocal or kit, and parallel vs serial compression compares them side by side.

Common parallel compression mistakes

Judging the blend while it's louder

Adding a compressed copy always makes the track louder, and louder nearly always sounds better for a few seconds. Turn the group down when you bring the bus in, or use the lab's Original and Compressed switch, so you're comparing the shape of the sound at the same level.

The two paths falling out of time

If the copy arrives even a millisecond late, it combines with the dry signal into a comb filter: a hollow, phasey tone with a row of notches through the spectrum. Your DAW's delay compensation normally lines the paths up, even when a plug-in adds latency, so check it's switched on if the blend suddenly sounds thin. Sending the copy out through hardware adds a round-trip delay too, and the DAW needs to know about it.

A slow attack on the copy

A slow attack is the usual advice for punchy compression in series, and it's the wrong habit here. It lets the transients into the copy, the makeup gain boosts them, and the peaks of the blend end up higher than the dry sound's. Keep the copy fast and let the dry path carry the punch.

Too much of the copy

Once the copy is loud enough to set the level of the peaks, you've lost the point of doing it in parallel. If the drums start to sound small and flat again, pull the bus down until the hits come back.

Squashing the dry path as well

Light compression on the dry track is fine. Heavy compression there takes away the transients the blend depends on, so the copy has nothing to sit under.

Train your ears

The quiz under the lab hides the compressor settings. Most rounds play the loop with none, 40% or all of a heavily squashed copy mixed in and ask how much of it you're hearing. The levels are matched, so listen to how dense the quiet parts are and whether the hits still jump out. The other rounds ask whether the attack is fast or slow, which decides whether the copy lets the front of each hit through. Your best score is saved in your browser.

Questions people ask

Keep going