Most engineers meet parallel compression as a recipe. Send drums to an aux, slam a 1176 with twelve dB of gain reduction, blend underneath the dry kit, listen to it sound bigger. The recipe works, but the reason it works is more interesting than the recipe itself, and it predates the New York studios that gave the technique its nickname by more than a decade.
The psychoacoustic asymmetry
The whole trick rests on one fact about hearing. The ear is highly sensitive to loud sounds suddenly getting quieter. It is much less sensitive to quiet sounds getting louder. Conventional downward compression, the kind we mostly use, attacks peaks and pulls them down. That is the side of the asymmetry the ear notices most, which is why heavy downward compression often sounds like it is squashing something even at moderate ratios.
Parallel compression works the other way around. You sum two signals. One is the dry source, untouched. The other is a heavily compressed copy of the same source. When a peak arrives, the compressor on the wet path clamps it hard, so the wet contribution shrinks to almost nothing and the dry transient dominates the sum. When the signal drops between peaks, the compressor on the wet path lets through a near full level copy, and that copy adds to the dry, lifting the soft passages. The mathematical outcome looks a lot like upward compression. Peaks roughly unchanged, softs raised.
Bob Katz coined the modern term parallel compression and framed it explicitly as a practical implementation of upward compression. The reframing matters, because once you see it that way you stop trying to listen for the wet bus on its own and start judging it by how much it lifts the body of the signal without dulling the attack.
Where it actually came from
The technique was not invented in New York. It was published in October 1977 by Mike Beville in Studio Sound magazine, in an article called Compressors and Limiters that was mainly about classical music recording. A separate section at the end of the article describes placing a limiter compressor in parallel with the direct signal to achieve effective compression at low input levels. The framing is exactly the lift the softs framing, twenty years before pop and hip hop producers turned it into a punchy drum trick.
It is even older than that, conceptually. The internal signal flow of Dolby A noise reduction, introduced in 1965, already routed audio through parallel buses with a compressor on one of them, mixed back in a flexible ratio. The Dolby system was built for noise reduction, not punch, but the topology is the same.
The New York name attached itself in the 1980s and 1990s when studios in the city built a reputation for relying on the technique for drums and vocals. They did not invent it. They popularised it.
One footnote worth nailing down. A lot of citations of the Beville article describe the technique as side chain compression. Beville did use the term side chain in the article, but he used it to describe the internal electronics of a compressor, not the parallel routing. The misattribution is common and worth ignoring.
Why the wet bus needs to be aggressive
If the wet compressor barely does anything, summing it back to the dry signal just makes everything a bit louder. You want a real difference between paths.
Typical parallel settings are far more aggressive than you would ever use as a normal insert. Ratios of 10:1, 20:1, even 50:1 are normal. Threshold is set well below the lowest signal level so the compressor is always working. Sound on Sound measured a configuration with a 50:1 ratio and a threshold of -23 dBFS and found it applied just over 20 dB of gain reduction as the input approached 0 dBFS. Stacking multiple parallel compressors keeps lifting the floor further. A single parallel compressor adds roughly 6 dB of perceived lift. Three stacked add roughly 12 dB.
This is also why the choice of compressor matters more in parallel than in conventional inserts. The character of the wet path is unmasked when it sums with the dry signal, so harmonic distortion and program dependent envelopes from classic units like the 1176, Distressor, Fairchild, and LA-2A become tonal contributions rather than artefacts you try to hide. Michael Brauer's well known vocal chain leans on this, blending several different parallel compressors with different colours into a single composite vocal bus.
Practical starting points
These are starting points, not absolutes, and every source disagrees by a few milliseconds.
For a drum bus, route the kit to an aux, insert an aggressive compressor, aim for 10 to 12 dB of gain reduction with a release around 100 ms. A common tip from Danny Echevarria is to mix the compressed bus around 6 dB below the dry drums to keep cymbals from pumping audibly.
For a lead vocal, a smoother optical or vari mu style compressor sits better than a FET. Try 5 to 10 dB of gain reduction with a 4 to 6 ms attack and a fast release, blended subtly underneath the dry vocal.
For bass, two parallel paths can work well. One with light 2 dB GR fully wet, contributing body. One aggressive at 5 dB GR blended around 40 to 50 percent, contributing consistency.
In mastering the picture changes. Most mastering engineers either avoid parallel compression or use it very lightly. 10 dB of gain reduction on the wet path is already considered a lot, 20 dB is generally too much, and a 50 percent wet to dry blend is a reasonable starting point you adjust by ear.
The digital pitfall
This is where parallel compression goes wrong in DAWs more often than people admit. Plugin compressors typically introduce somewhere between 0.3 and 3 milliseconds of latency, more if they use lookahead. If the wet path is even a millisecond behind the dry path when they sum, the two add together as if you had a short delay line between them. The result is comb filtering, narrow nulls and peaks scattered across the spectrum, audible as a hollow or phasey quality on the summed signal.
Modern DAWs handle this automatically for aux sends by inserting plugin delay compensation upstream of the dry path. It works, but you should check. Bypass the wet path and the dry path should be unchanged in level and tone. If the dry tone shifts when you toggle the wet bus, latency compensation is not doing its job.
This is also why compressor plugins with built in mix or wet/dry knobs have become so popular. FabFilter Pro-C 2, the Ableton stock compressor, and many others perform parallel compression internally on a single channel, with zero delay between the dry and wet paths. The trade off is that you give up the ability to choose a different compressor for the wet path. For drums and vocals where the colour of the compressor matters, the aux send approach still wins. For utility duties on individual tracks, the mix knob is cleaner.
One more routing gotcha. The aux send feeding the parallel compressor should be pre fader, not post fader. With a post fader send, every move of the source channel's fader changes how hard the compressor on the wet path hits, so the blend keeps shifting under you. Pre fader fixes the level the compressor sees, and the channel fader only affects the dry contribution.
What it actually buys you
Parallel compression is sometimes described as cheating, as if it lets you have compression without the side effects. It is not quite that. What it buys you is the ability to lift the body of a signal without softening its attack, because the attack is preserved by the untouched dry path. Whether the result actually sounds better depends on the source. On a drum kit that is already balanced and just needs weight, it is hard to beat. On a vocal that already sits well in the mix, a heavy parallel bus can make it feel less defined rather than more. The technique is a tool with a specific job, lifting softs, and like any tool it is worth being honest about whether that is the job you actually have.