Parallel vs serial compression

At a glance

What's the difference between parallel and serial compression?

The difference is where the compressor sits. In serial compression every compressor is in the path of the sound, one after another, so what you hear has been through all of them. In parallel compression the sound splits in two. One path goes through a compressor set much harder than you’d ever use on its own, the other stays exactly as it was, and the two are mixed back together.

That one change decides what happens to a transient, the sharp spike at the very start of a drum hit or a sung consonant. In serial, every compressor gets a chance to turn it down. In parallel, the dry path carries it through untouched, so the hit keeps its shape however hard the copy is squashed. What the squashed copy adds is everything quieter: ghost notes, room sound, the tails of notes and the ends of words.

Both reduce dynamic range, the gap between the loud and quiet moments. Serial gets there by turning the loud moments down. Parallel gets there by turning the quiet moments up, which is why it behaves like upward compression. If the controls themselves are new to you, what is compression? covers threshold, ratio, attack and release from the start.

Reading the lab

The lab has two compressors and four ways to wire them, picked with the cards along the top. One puts a single compressor in the path. Serial puts comp 1 and comp 2 in a row. Parallel takes comp 1 out and puts comp 2 on a copy, blended with the dry sound by the mix knob, and Both keeps comp 1 in the path with a parallel copy after it.

The wide screen is the signal chain. Each box shows the whole loop at one point: the pale shape is what went into that box and the coloured one is what came out, with a gain reduction lane under each compressor. In the parallel routings you can see the dry line running over comp 2 to the mix point. Every compressor gets automatic makeup gain that matches it to the loudness of what went in, and the output is matched to the original loop, so switching routings never makes anything louder. Louder nearly always sounds better for the first few seconds, so a fair comparison needs the levels matched.

Underneath, the close-up shows one hit, original against output at the same loudness, and reads out how far the loudest 10 ms of it has moved. The transfer curve is the whole chain’s, and its corner label says how hard the chain squeezes the top 12 dB of input. The panel on the right explains what your settings are doing with numbers measured from the loop.

The same drums, both ways

The first three steps of the lab run the drum loop through one compressor and then through a parallel copy, with numbers you can check on the panel as you go.

One compressor at −24 dB and 6:1 with a 0.3 ms attack takes up to 18 dB off the loudest hits. With the level matched to the original, the quietest tenth of the loop comes up by about 11.5 dB, which is a lot of extra detail. The cost is the hits. The loudest 10 ms of the loop drops by about 3 dB relative to everything else, and the close-up shows why: the compressor lets through a spike a fraction of a millisecond long before it clamps down, and then flattens the body of the hit behind it.

Now take that compressor out of the path and put a harder one on a copy, at −40 dB and 10:1 with a 0.1 ms attack, blended in at 30%. The copy is turned down by more than 30 dB on the loudest hits, which would sound awful on its own. Mixed under the dry drums, it lifts the quiet parts by about 12 dB, slightly more than the single compressor did, while the loudest hit moves by only about 1 dB.

Push the mix to 70% and the copy starts to set the shape of the hits as well. The quiet parts come up by 18 dB, and the loudest hit drops about 3 dB again, close to where the single compressor left it. A parallel blend pushed far enough turns back into ordinary heavy compression.

Why put two compressors in a row?

On a held level, two compressors in a row with the same threshold add up to one compressor with a bigger ratio. Say you have two at a −20 dB threshold and a 2:1 ratio, with no makeup between them, and a peak arrives at −4 dB. The first compressor lets half of its 16 dB overshoot through, so the peak leaves at −12 dB. That’s 8 dB over the second threshold, and at 2:1 it comes out at −16 dB, exactly where a single 4:1 compressor would have put it. With matching thresholds and no makeup gain between them, ratios in a row multiply.

So the reason to split the job is how each compressor gets there, and that comes down to timing. A single compressor has one attack and one release, and a vocal needs two kinds of control at once. The odd word that jumps 10 dB needs catching in a few milliseconds. The slow drift between a quiet verse line and a loud chorus line needs a gentle hand that moves over hundreds of milliseconds. Any single setting is a compromise: fast enough for the words and it pumps on the phrases, slow enough for the phrases and the loud words slip past.

In series, each compressor can have the setting that suits its job. Steps 4 and 5 of the lab do this on the vocal. One compressor at −26 dB and 6:1 takes up to about 16 dB off the loudest words and shrinks the loud to quiet gap from about 13 dB to 5 dB. Split it into a fast 4:1 compressor with a 0.5 ms attack, followed by a slow 2.5:1 one with a 15 ms attack and a 300 ms release, and the first takes up to about 9 dB while the second takes up to about 7 dB. The gap ends up near 6 dB, close to the single compressor’s, and neither compressor is working hard. In the chain screen, comp 1’s lane jumps on single words while comp 2’s rises and falls with the phrases.

The two stages can have different characters as well as different timing. A common vocal pairing is a fast FET compressor like the Universal Audio 1176 catching the peaks, followed by an optical one like the Teletronix LA-2A doing the slower levelling, and the plug-in versions of both work the same way in a DAW. Opto vs FET vs VCA vs vari-mu explains how those designs differ.

Serial and parallel on a transfer curve

Lay the two routings on one graph and you can see where each one works. Serial compression squeezes hardest at the top, because the ratios stack up there. Parallel compression squeezes hardest in the middle; above the point where the copy comes out quieter than the dry sound, the dry sound takes over again and the curve runs almost parallel to the diagonal.

In the figure, with both matched at the peak, an input at −60 dB comes up by 15 dB through the serial pair and by about 14 dB through the parallel blend, so at the very bottom they do much the same job. Higher up they part company. Through the serial pair the top 20 dB of input is squeezed into 5 dB of output, while the parallel blend lets the top 12 dB of input through as about 10.7 dB. The lab’s transfer curve shows the same thing for whatever settings you’ve dialled in: its label reads about 10:1 across the top for the serial vocal settings and about 1.2:1 for the parallel ones.

The curves are drawn for held levels, which leaves out attack and release, but they still explain why parallel compression is the gentler of the two on peaks, even when the copy is flattened by 30 dB.

Can parallel compression control a vocal?

Not on its own. Try the parallel copy on the vocal at step 6: it fills in the quiet syllables and the ends of words, but the loud to quiet gap only comes down to about 9 dB. The dry path still carries every loud word at its full size, so in a dense mix those words jump out as much as they did before. If a part’s level wanders, it needs a compressor in its path.

Which one should you use?

Ask what’s wrong with the sound. If the loud moments jump out or the level drifts, you need control, and that’s a job for serial. If the level is fine but the part sounds thin, or the quiet detail disappears in the mix, you need density, and parallel adds it without costing you the transients.

On drums, parallel is the usual first move because their loud and quiet parts are so far apart: a snare hit can peak 20 dB or more above the ghost notes around it. The parallel compression lesson covers setting up the bus in detail, and the drum bus lesson covers what usually goes on the drum group alongside it.

Using serial and parallel together

You can use both on the same part. A common vocal chain has a compressor in the path to catch the loudest words, then a send from after that compressor to a parallel bus that adds density underneath. Because the send comes after the first compressor, the copy works on a vocal that’s already under control, so it doesn’t over-react to the loudest words.

Step 7 does this. Comp 1 sits in the path at 3:1 and the parallel copy runs after it at 25%. The gap comes down to about 6 dB and the quiet parts come up about 5 dB, more lift than serial or parallel managed alone on this vocal.

How to set up serial compression

Parallel setup has its own walkthrough on the parallel compression page. This is the serial version on a lead vocal.

1. Put two compressors on the trackAny two will do. A fast one first and a slower or smoother one second is the usual order.
2. Bypass the second and set up the first as a peak catcherUse a 4:1 to 8:1 ratio and an attack under 1 ms, and lower the threshold until only the loudest words show 3 to 6 dB of gain reduction. Quieter words shouldn’t move the meter at all.
3. Match its levelAdd makeup gain until bypassing it makes no difference to loudness. This also keeps the level going into the second compressor where it was.
4. Switch on the second and set it up as a levellerUse a 2:1 to 3:1 ratio, an attack of 10 to 30 ms and a release of 200 ms or more, and lower the threshold until it shows 2 to 4 dB on most phrases. Its meter should move slowly.
5. Match its level too, then compareBypass both at matched loudness. If you can’t hear an improvement, ease the thresholds back up before adding anything else.

If you change the first compressor later, check the second. Its threshold is measured on what the first one sends it, so a change in the first one’s makeup gain moves how hard the second one works.

Common mistakes

Comparing at different levels

Both routings make a part louder once the makeup gain is in, and adding a parallel copy makes it louder still. Turn the result down to match before you decide anything; the lab does that for you.

Two compressors doing the same job

Two compressors in a row with the same attack and release behave much like one compressor with a higher ratio. Serial compression pays off when the stages are different, so give each one a job of its own.

A slow attack on the parallel copy

A slow attack lets the front of each hit into the copy, the makeup gain boosts it, and the peaks of the blend end up higher than the dry sound’s. That eats headroom, and the copy adds snap where you wanted body. If you’re after density, keep the copy fast and let the dry path carry the transients.

Expecting parallel to control a vocal

A parallel copy adds weight under the loud words and leaves them as loud as they were. If the level jumps around, put a compressor in the path.

A parallel bus out of time

If the copy arrives even a millisecond late, it adds to the dry sound as a comb filter, which sounds hollow and phasey. Your DAW’s delay compensation normally lines the paths up, so check it’s switched on if the blend suddenly sounds thin. A serial chain can’t have this problem, because there’s only one path.

Training your ears to hear the difference

The quiz under the lab plays the drum loop either through two compressors in series or through a crushed copy in parallel, at the same loudness and with the controls covered, and asks which. Listen to the shape of each hit. In serial the body of the hit is pressed down and only a thin click gets through at the very front; in parallel the hit keeps its own shape, with the room filled in behind it. The last four rounds ask how much of the copy is mixed in, which is the setting you’ll adjust most on a real parallel bus.

Questions people ask

Keep going