Compression ratio, threshold, and knee

The short answer

You already know what a compressor does: it turns loud parts down so the gap between your loudest and quietest moments shrinks. Three controls decide exactly how that happens, and none of them are about timing. The threshold sets the level where compression starts. The ratio sets how hard it clamps down once you cross that level. The knee sets how abruptly or gently it engages right around the threshold.

These are the static, or level, controls: hand a compressor a steady input at a given level and these three alone decide the output level. The time controls, attack and release, decide how fast it gets there, and they get their own deep dive. Here we stay with the picture that shows threshold, ratio, and knee all at once.

One picture holds all three

The most useful way to see a compressor is not a knob panel but a graph: input level along the bottom, output level up the side, both measured in decibels below full scale. Every input level maps to exactly one output level, so the compressor becomes a single line, called the transfer curve. A processor that did nothing at all would be a straight diagonal: whatever goes in comes out the same, output equals input. That 1:1 diagonal is the reference every compression curve bends away from.

Read the curve left to right and you are reading quiet to loud. The left half sits right on the diagonal, which is the compressor telling you it is doing nothing to your normal speaking level. Somewhere in the middle the line breaks away and tilts flatter. Everything interesting about a compressor happens at and after that break.

Threshold: where compression starts

The threshold is the input level where the curve leaves the diagonal. Below it, nothing is touched. Above it, gain reduction kicks in. So the threshold is really a decision about how much of your signal gets compressed at all. Set it high, close to your peaks, and only the loudest moments are affected while the body of your voice passes clean. Drag it down toward your average level and more and more of your speech crosses the line, so the compressor works on nearly everything and the overall effect gets stronger.

This is why threshold and input level are two sides of one coin. Turning the threshold down by 6 dB and turning your input up by 6 dB push the same amount of signal over the line. If a compressor suddenly sounds too aggressive after you boost a gain stage earlier in the chain, the threshold did not move; your signal climbed past it. Getting your gain staging right first is what makes a threshold setting mean something stable.

Ratio: how hard it clamps

Once the signal is over the threshold, the ratio decides how much of the overshoot survives. It is the slope of the line above the break, written as a comparison like 4:1, which reads as four to one. At 4:1, for every 4 dB the input goes over the threshold, the output is allowed to rise only 1 dB. The other 3 dB become gain reduction. The math is exactly that simple:

On the plot this is the fan of lines spreading out from the threshold. A low ratio like 2:1 barely tilts off the diagonal, a gentle leveling you can lean on for a whole voice. A high ratio like 10:1 lies almost flat, clamping hard so almost none of the overshoot gets through. Push the ratio higher still and the slope keeps flattening. In the limit, an infinite ratio would let none of the overshoot through: the line goes perfectly horizontal and the output simply cannot rise past the threshold. That flat wall is exactly what a limiter is, which is why a limiter is just a compressor with the ratio pinned to the extreme, and why the practical split between the two is a matter of degree, not kind.

Knee: how gradually it engages

The threshold and ratio describe two straight lines meeting at a corner. The knee decides whether that meeting is a sharp bend or a smooth curve. A hard knee is a knee of zero: the compressor does nothing right up to the threshold, then snaps to the full ratio the instant you cross it, a crisp corner on the plot. A soft knee rounds that corner off over a range of input levels, easing the ratio in gradually as the signal approaches the threshold and only reaching the full ratio some decibels above it.

In the graphic, the dashed elbow is the hard-knee corner, and the bright 4:1 curve is the same setting with a soft knee rounding it. The rounding is subtle by nature, a fraction of a decibel of extra reduction spread across the transition, but it changes how the compression sounds. A hard knee makes the onset of gain reduction easy to hear, which is useful when you want obvious control or a tight grip on peaks. A soft knee starts working before the level fully arrives, so the compression fades in instead of switching on, which tends to sound more natural and transparent on a voice. Neither is correct; a soft knee hides the work and a hard knee announces it.

Reading the three together

Put them in order and any compressor becomes legible. First find the threshold, the input level where the line leaves the diagonal, and you know which part of your signal is affected. Next read the slope above it, the ratio, and you know how hard the affected part is squeezed. Finally look at the corner, the knee, and you know whether that squeeze arrives abruptly or eases in. A gentle vocal leveler is a low threshold that catches most of the voice, a modest 2:1 or 3:1 ratio, and a soft knee so the work stays invisible. A firm grip on peaks is a higher threshold, a steep ratio, and a harder knee.

One thing this picture deliberately leaves out is time. The transfer curve assumes a steady input, but real speech is transients and tails, and how quickly the compressor climbs to the reduction the curve promises, then lets go again, is the job of attack and release. Threshold, ratio, and knee decide how much; attack and release decide how fast. You need both to describe what a compressor actually does to a moving signal.

In patchd

In patchd the Compressor is a free node you add to any channel, and its level controls map straight onto the curve above. Threshold runs from -60 dB up to 0 dB and defaults to -18 dB, the break point drawn in the graphic. Ratio runs from 1:1 (no compression) up to 20:1 (a hard clamp) and defaults to 3:1, a natural leveling ratio that sits between the 2:1 and 4:1 lines shown. Knee runs from 0 dB (a hard corner) to 24 dB (a very gradual, rounded onset) and defaults to 6 dB, the soft knee in the plot.

A couple of honest notes. The Compressor is an inline node, so it processes the signal in place and has no wet/dry mix control. It also has a Makeup Gain to bring the level back up after reduction, plus a sidechain high-pass and low-pass that filter only the compressor's detection signal, not your audio, so heavy bass does not trigger reduction on the whole voice while the sound itself still passes full range. When you need the ratio taken all the way to a wall, that is a job for the separate free Limiter node rather than the Compressor.

patchd is a Windows audio mixer in development, with these dynamics nodes free on every channel and the real-time engine identical on both the Free and Studio tiers. If you want a mixer where reading a compression curve translates directly into three honest controls, join the waitlist and we will tell you the moment it is ready to install.