How to set up a compressor on your voice

This guide uses the compressor in patchd, a free effect on every channel, and its settings are organized exactly the way the guide is: the node panel groups its controls into beginner, intermediate, and advanced, so you can stop at any tier and have a working compressor. If you want the concept first, start with what a compressor is, then come back here to dial one in.

The four settings that matter

A compressor has one job: when your voice gets louder than a line you draw, turn it down by an amount you choose, quickly, and then let go. Four controls decide the line, the amount, and the timing.

• Threshold (default -18 dB). The loudness line, and the only setting you tune against your own voice. Speak normally and watch the live meter: your conversational level should sit just under the threshold, and only your laughs, shouts, and leaned-in moments should cross it. If the compressor is working on every single word, the threshold is too low; if it never moves, it is too high.
• Ratio (default 3:1). How hard the peaks get turned down: at 3:1, every 3 dB you go over the threshold comes out as 1. The rules of thumb: 2:1 is gentle (10 dB over becomes 5 over), 4:1 is solid for voice, and 10:1 or higher stops being compression and starts being a limiter.
• Attack (default 10 ms). How fast the compressor clamps down once you cross the line. Around 10 ms is the sweet spot for speech: fast enough to catch the body of a loud word, slow enough to let the crisp front of your consonants through untouched.
• Release (default 100 ms). How fast the gain comes back once you quiet down. Around 100 ms follows the natural rhythm of speech. Much faster and the level audibly pumps between words; much slower and one loud sentence drags the next one down with it.

Start with the defaults, speak the way you actually will on stream, and move only the threshold until the loud moments get caught and the normal ones pass clean. For most voices that is the entire setup. Everything below this line is refinement.

Knee and makeup gain

The intermediate controls fix the two ways a basic compressor gives itself away: an audible grab the instant you cross the threshold, and a mix that got quieter overall because you turned the peaks down.

• Knee (default 6 dB). How gradually compression fades in around the threshold. At 0 dB the knee is hard: nothing happens at -18.1 dB and full-ratio compression happens at -17.9, which you can hear as a grab on every peak. At the default 6 dB the compressor starts easing in 3 dB below the threshold and reaches full ratio 3 dB above it, so the transition is a curve instead of a corner. Soft knees in the 6 to 12 dB range sound more natural on voice; save the hard knee for when you want the clamp to be obvious.
• Makeup gain (default 0 dB, up to 24). Compression only ever turns things down, so a compressed voice is a quieter voice. Makeup gain adds a fixed boost after the compression to restore the loudness the peaks gave up. Toggle the node on and off while speaking and raise makeup until both versions feel equally loud; now you have the same level with the spikes shaved off, which was the point all along.

The math, straight from the engine

The engine computes one number per instant: how many dB of gain reduction to apply, given the detected level. The curve has three regions, and their boundaries come from your threshold T and knee width:

Below kneeStart the gain is simply 0 dB: the compressor does nothing, which it is remarkably good at. Above kneeEnd, the reduction grows linearly with how far over the threshold you are:

Inside the knee, the engine blends between those two behaviors with a quadratic:

Run the defaults through it. T is -18 and R is 3, and a shout peaks at -6 dB: that is 12 dB over the threshold, so the gain is (1/3 - 1) * 12 = -8 dB, and the peak comes out at -14 dB, just 4 dB over the line. Twelve over in, four over out: whatever exceeds the threshold gets divided by the ratio, which is what 3:1 was promising on the label.

One detail worth knowing: the curve drawn in the patchd node panel is generated from the same formula the audio path runs, kept in lockstep by construction. The plot is not an artist's impression of the compressor; it is the compressor. What you see is what you hear, by construction.

The remaining advanced controls shape what the compressor listens to, not what it does. The sidechain high-pass (20 Hz to 2 kHz) and sidechain low-pass (1 kHz to 20 kHz) filter the detection signal only, never your audio. Raise the high-pass so that low-frequency thumps and plosives stop triggering gain reduction on your whole voice; at the default 20 Hz and 20 kHz endpoints both filters are bypassed and the detector hears everything.

Where the compressor sits in the chain

Order matters, because every effect feeds the next. The compressor goes after your cleanup, always: high-pass or EQ first, noise suppression next, then the noise gate, and the compressor after all of them. The reason is the makeup gain you just set: a compressor narrows the distance between your quietest and loudest sounds, and if the room noise is still in the signal, that noise gets pulled up toward your voice level right along with your whispers. Compress first and you hand the gate a raised, unstable noise floor and a threshold that no longer means anything. Clean first, then even out what is left.

In patchd the compressor is a free node in the effects rack on every channel. Drag it in after the gate, set the threshold against the live meter, and the evened-out result is what every destination hears: your headphones, your stream bus, all of it. patchd is in development now. Join the waitlist and we will tell you the moment it is ready to install.