Audio Compression Explained (Threshold, Ratio, Attack, Release)

A compressor turns the loud parts of a recording down. That's the whole machine. Imagine an engineer with one hand on your channel fader, pulling it down every time you get loud and letting it back up when you get quiet, except the hand reacts in milliseconds and never gets bored. This is audio compression explained from that one idea: threshold decides when the hand grabs, ratio decides how hard, attack decides how fast, and release decides when it lets go. I'll teach it on the Compressor in SoundCraft, our free browser DAW, because its panel puts a live gain-reduction meter front and center, and that meter is how you actually learn this effect.

Why compression is hard to hear (and why that's fine)

Reverb announces itself. EQ you can hear in two seconds. Compression is different: when it's working well, the track just sounds a bit more even, a bit more finished, and you can't point at why. Beginners twist the knobs waiting for an obvious change, hear nothing, and crank everything until the sound pumps like a broken radio. The fix is to stop listening for the effect and start watching the gain-reduction meter, the orange strip labeled GR in SoundCraft's panel. It shows, live, how many dB the compressor is turning you down at this exact moment. Once you trust the meter, every control makes sense.

Audio compression explained: the four controls

Click a clip so its channel is focused, open the ☰ effects rack, and add the Compressor. Here's the panel fresh out of the box, mid-playback on a vocal take that swings between quiet and belted phrases:

• Threshold (-60 to 0 dB) is the level where the compressor wakes up. Anything under it passes through untouched; anything over it gets turned down. In SoundCraft the threshold is a white caret you drag directly on the input meter, so you can literally place it against the signal you're watching. Arrow keys move it 1 dB at a time (Shift jumps 10).
• Ratio (1:1 up to an ∞:1 detent) is how hard the hand pushes. At 4:1, a peak that goes 8 dB over the threshold comes out only 2 dB over. Higher ratios squash harder; the ∞:1 position stops peaks almost dead, which is limiting rather than compression.
• Attack (0.01 ms to 1 s) is how fast the hand grabs once a peak crosses the line. This one shapes the character more than any other knob, and we'll spend a whole section on it below.
• Release (1 ms to 1 s) is how fast the hand lets go after the peak passes. Too fast and the level flutters. Too slow and the compressor never recovers between words, so everything stays ducked.

Two more sit on the panel. Output (-12 to +24 dB) is makeup gain, a second caret riding the output meter; since compression only ever turns things down, you add gain back here to match the original level. And Mix blends the compressed signal against the dry one, which turns out to be a big deal (more on that at the end). One quirk worth knowing: the knee is fixed at a gentle 6 dB, so the compressor eases into reduction rather than snapping at the exact threshold. And the Release knob tops out at 1 second, a hard cap in the browser audio engine, which in practice you'll never miss on vocals or drums.

A vocal setting you can copy

1. Create a free account. The editors cost nothing; the paid plans sell storage and server compute, not effects.
2. Open SoundCraft and drop in a vocal take (a temporary session works).
3. Click the clip, open the ☰ rack, add the Compressor, and hit play.
4. Drag the threshold caret down until the GR meter shows 2 to 4 dB on the loud phrases and nothing on the quiet ones.
5. Set Ratio to 3:1, Attack around 15 ms, and add makeup gain to taste.

Compare the two screenshots. Same singer, same phrase. At the defaults the meter buried itself near 13 dB of reduction, which flattens a performance into cardboard. With the threshold raised and the ratio eased, the compressor only trims the peaks that actually stick out, and the take keeps its shape. That difference is invisible on the knobs and obvious on the meter. If your take needs cleanup before this stage, our guide to recording vocals in the browser covers getting a usable performance in the first place.

Attack is the punch knob

Every drum hit, consonant and pluck starts with a transient, a few milliseconds of spike that carries all the impact. The attack time decides whether that spike survives. A fast attack (under about 5 ms) clamps before the transient finishes, which rounds it off and pushes the sound backwards. A slower attack (20 to 40 ms) lets the spike through untouched and then clamps the body of the note behind it.

Here's the counterintuitive part: on drums, a slower attack makes things hit harder. The transient passes clean, the sustain gets pulled down, and the contrast between them grows. That contrast reads as punch. So if a snare feels soft, don't reach for an EQ boost first; try a compressor at 4:1 with a 30 ms attack and watch what the extra snap does. The reverse trick works too: a vocal with spitty, popping consonants calms down with a faster attack, since the compressor catches each consonant before it jumps out.

Release and the pumping sound

Release is timing. After a peak, the compressor climbs back to zero reduction at the speed you set, and you can watch it happen on the GR strip. The goal for most material: the meter should recover to zero right before the next loud moment arrives. On a vocal, that means it resets between phrases. On a drum groove, it bounces with the beat.

Get it wrong and you hear pumping, the whole track audibly swelling as the compressor lets go. Dance music uses that on purpose (sidechain pumping is a genre sound at this point), but on a vocal it reads as broken. SoundCraft's 250 ms default release suits a mid-tempo song; shorten it for fast material, lengthen it if you can hear the level breathing. There's no formula better than the meter and your ears together.

The glue trick, and when to skip compression entirely

Compression also works as a finishing move. Select your drum channels, press Ctrl+G to group them into a bus (same move as in the reverb post in this series), and put one Compressor on the group at 2:1 with a slow attack and only 1 to 3 dB of reduction. The kit starts moving as one instrument instead of five separate recordings. Engineers call it glue, and it's subtle enough that you should toggle the enable button to confirm it's real.

And sometimes the right amount is none. A lot of AI-generated songs arrive already mastered and dense, with almost no dynamic range left; compressing them again just adds distortion. Programmed synths that never vary in level have nothing for a compressor to do. And a take that's consistently too quiet needs a fader, not a compressor, because turning down peaks that don't exist accomplishes nothing. If you can't name the problem you're fixing, leave the compressor off. An EQ cut solves tone problems; compression only solves level problems.

One knob most hardware never had

The Mix knob defaults to 100%, fully compressed. Drop it to around 50% and you get parallel compression: the dry signal at full strength with a squashed copy blended underneath. This is the classic New York drum trick, done on hardware with a spare bus and careful routing, and here it's one knob. Try it with a deliberately heavy setting, 8:1 and a low threshold, then blend to taste. The transients stay because the dry path carries them; the squashed copy fills in the quiet detail underneath.

The short version: set the threshold against the meter so it only catches the loud moments, keep ratios modest (3:1 vocals, 2:1 on a bus), slow the attack when you want punch, time the release so the meter resets between hits, and match loudness before you believe your ears. Sign up free and run a real take through it; twenty minutes of watching the GR meter move will teach you more than every diagram on the internet.