Audio foundations

Compressor vs limiter

A limiter is a compressor pushed to its mathematical extreme, which is exactly why people confuse them. One shapes your voice all day; the other stands by a ceiling and does nothing until the one moment it matters. This page draws the line, with the real numbers.

If you want either tool explained from zero, start with what a compressor is or what a limiter is. This page is the comparison: the same math at two different settings, the numbers each one ships with, and the slot each one has earned in a voice chain.

Beginnerstart here

One machine, two extremes

A compressor turns down whatever crosses its threshold, by a ratio. At its default 3:1, every 3 dB you go over the line comes out as 1 dB over. That is a push-back, not a wall: get louder and the output still gets louder, just a third as fast. It engages on most of your speech and shapes it gently, which is what you want from a tool that runs all day.

Now push the ratio up. At 10:1 a compressor already behaves like a limiter, and as the ratio heads toward infinity the slope above the line flattens to nothing: no matter how far over you go, the output stays put. That flat line is a limiter, which is why the extreme version is called a brickwall. A difference in kind falls out of a difference in degree. A compressor negotiates with your peaks; a limiter refuses them.

Reach for the compressor when the complaint is about consistency: quiet words vanish, loud moments jump out, the level wanders as you turn your head. Shaping is constant work across the whole performance, and the compressor is built to do it without being heard.
Reach for the limiter when the complaint is the rare disaster: the clipping crackle on a big laugh, the shout at a jump scare, the mic bump. Safety is a job with nothing to do 99 percent of the time, and that is the point.

compressor 3:1limiter -1.0 dBno processing

Same input, three outcomes. Below their thresholds all three paths ride the same diagonal; the compressor bends away softly at -18 dB and climbs at a third of the rate, the limiter holds the diagonal until -1.0 dB and goes dead flat, and only the unprocessed signal keeps climbing into the clipping zone.

Intermediategoing deeper

Shaping vs safety, in real numbers

The two nodes ship with defaults that spell out their jobs. The compressor sits at a threshold of -18 dB, low enough that ordinary emphasis crosses it, with the 3:1 ratio, a 6 dB soft knee that rounds the corner so the push-back fades in gradually, a 10 ms attack that lets the front of each word punch through, and a 100 ms release that rides your pauses. Run the default math on a real peak: a word that hits -6 dB is 12 dB over the line and gets pulled down 8 dB, coming out at -14 dB. It worked, and you barely heard it work.

The limiter's defaults describe the opposite temperament. The ceiling sits at -1.0 dB, a level a well-compressed voice should never reach, with a 0.5 ms attack that clamps a runaway peak mid-syllable and a 50 ms release that lets go the moment it passes. On a well-set chain the limiter does nothing at all for minutes at a time; its entire value is the one moment the compressor was never set up for. The ceiling defaults to -1.0 rather than 0 to leave headroom for inter-sample peaks, a story told in full in what a limiter is.

One design fact separates patchd's limiter from the mastering limiters you may have read about. Mastering limiters buy perfect transient catching with lookahead: they delay the audio by 1 to 5 ms so the clamp lands before the peak arrives, and that delay is added latency for everything passing through. patchd's limiter is deliberately zero-lookahead, so it adds zero latency and stays safe in a path you are monitoring live. It accepts a slightly less perfect grab of the first instants of a transient in exchange, and for a live voice that is the right trade.

Advancedthe math

The same formula, two settings of R

Above the threshold, both behaviors come out of one line of engine code, the same formula that draws the curve in the node panel:

gain_dB = (1/R - 1) * (in_dB - T)

the engine's static gain, above the knee

At the compressor's defaults (T = -18 dB, R = 3), a -6 dB peak is 12 dB over: gain = (1/3 - 1) * 12 = -8 dB, so it comes out at -14 dB. Feed the same peak to the limiter's version of the formula and it comes out pinned at the ceiling.

The knee is the compressor being polite about the transition. For 6 dB around the threshold, the gain follows a quadratic instead of a straight line, chosen so the slope of the curve is continuous through the bend: no audible kink as a word crosses the line. The limiter declines the courtesy on purpose. At the ceiling its slope snaps from one to zero, because a wall that eased into being a wall would let the peak through first.

slope above the line = 1/R (3:1 -> 1/3, R -> inf -> 0)

the whole comparison in one number

Every dB over the line keeps 1/R of itself. The compressor hands a third of it back; the limiter hands back nothing.

That single number is the entire comparison. One-third is shaping: your dynamics survive, just smaller. Zero is safety: past the ceiling, dynamics stop existing.

Compressor mid-chain, limiter last

EQBYP

Noise SuppressionBYP

GateBYP

CompressorBYP

LimiterBYP

The compressor evens, the limiter guards the exit. Mid-rack, the compressor works on a voice the earlier nodes have already cleaned; the limiter takes the last slot, so nothing after it can push past the ceiling.

The compressor belongs in the middle of the rack, after the cleanup. An EQ and noise suppression in front of it remove energy its detector should never count, and the gate ahead of it matters even more: a compressor's makeup gain raises everything, including the room, and placed before the gate it can lift the noise floor right over the gate's threshold. The full reasoning, with the numbers, is in what order DSP effects should go in.

The limiter takes the final slot, and the logic is airtight: any node placed after it could push the level back over the ceiling and quietly cancel the guarantee. In patchd both are free nodes on every channel, added from the + Add Node menu, and the rack reorders by dragging, so putting them in this order is a ten-second job.

Which one do you need?

Name the complaint. If the voice itself is uneven, words disappearing and spiking from sentence to sentence, that is the compressor's job, and no ceiling will fix it. If the voice sounds fine until the one loud moment that crackles, that is the limiter's, and more compression would dull everything else to solve a once-a-stream problem. Most voice chains eventually run both: the compressor shaping every sentence, the limiter standing silently behind it as insurance.

In patchd the compressor and limiter are both free nodes in the effects rack, and each node's panel draws its curve from the same formula the engine runs, so the bend and the wall you saw above are the shapes your voice actually passes through. patchd is in development now. Join the waitlist and we will tell you the moment it is ready to install.