The short answer
A limiter is an audio tool that enforces a maximum level. You pick a ceiling, say -1 dB, and the limiter holds your peaks to it. Below the ceiling, your audio passes through untouched. When a peak crosses the line, the limiter pulls the gain down fast and keeps it down while the peak lasts. It has no lookahead, so the first few milliseconds of a sudden peak can get a dB or more over the line before the gain catches up; after that each cycle stays within a fraction of a dB of it. It is the last line of defense between your voice and digital clipping, the harsh crackle you get when a signal tries to be louder than digital audio can represent.
Conceptually, a limiter is a compressor taken to its logical extreme. A compressor turns loud moments down by a ratio: at 3:1, every 3 dB you go over the threshold comes out as 1 dB over. Raise that ratio and the output gets flatter and flatter above the line. Push the ratio toward infinity and the slope flattens into a ceiling: every dB over the line is turned back down to it. That flat top is a limiter.
In Patchd the limiter is a free node you add to any channel from the + Add Node menu, and it has exactly three controls:
- Ceiling (default -1.0 dB). The line your peaks are held to. Adjustable from -12 dB up to 0.
- Release (default 50 ms). How quickly the limiter lets go after the peak passes, so the moment after a loud word does not stay squashed.
- Attack (default 0.5 ms). How fast the limiter clamps when a peak hits the ceiling. Half a millisecond is fast enough that a shout gets caught mid-syllable.
Limiter vs compressor: safety vs shaping
Since a limiter is mathematically an extreme compressor, why run both? Because they do different jobs, and the jobs want different settings.
A compressor shapes your dynamics. It works on most of your speech, gently and musically: a moderate ratio, a soft knee, and an attack slow enough to let the natural punch of a word through. It is the tool that makes a voice sound even and produced, and the companion to EQ in any voice chain.
A limiter is safety. It should do nothing at all for 99 percent of your audio. Its whole purpose is the one moment the compressor was not set up for: the sudden laugh, the shout at a jump scare, the mic bump. A compressor set for natural speech will let a genuinely explosive peak through several dB hotter than everything else, and if that peak reaches 0 dB it clips. The limiter sits after the compressor and catches exactly those escapes. Compressor shapes, limiter protects.
That division of labor is why the limiter goes last among your dynamics processing, after the rest of the chain has done its work. Anything placed after the limiter could push the level back over the ceiling and undo that protection, and in Patchd that includes the strip fader, which comes after the whole rack: keep it at or below 0 dB if the ceiling is your safety net.
Setting the ceiling honestly
The obvious ceiling is 0 dB, the absolute digital maximum. The default in Patchd is -1.0 dB, and the gap is deliberate. A limiter measures the digital samples, but when those samples are converted back to an analog waveform, the smooth curve between two samples can swing slightly higher than either sample. These inter-sample peaks, called true peaks, can nudge past the sample values by a fraction of a dB. A ceiling of -1 dB leaves headroom for them, so a signal that measures legal stays legal all the way through playback. Streaming and voice platforms also run their own processing on your audio, and handing them a signal parked at exactly 0 gives their processing nowhere to work.
The math, and an honest tradeoff
Start with the compressor gain formula and push the ratio to the wall. A compressor above its threshold applies gain of (1/R - 1) times the overshoot. As R grows, 1/R falls toward zero:
gain_dB = (1/R - 1) * (in_dB - ceiling) -> ceiling - in_dB as R -> infAt R = 3, a peak 12 dB over comes out 4 dB over. As R approaches infinity, the same peak comes out 0 dB over: output pinned to the ceiling. An infinite ratio is just a compressor that stopped negotiating.
Now the tradeoff every limiter designer has to face. A limiter reacts to a peak, but even a 0.5 ms attack means the very first fraction of a transient can slip past before the clamp fully lands. Studio mastering limiters solve this with lookahead: they delay the audio by 1 to 5 ms so the detector sees each peak before the listener does, and the gain is already down by the time the peak arrives. Transient catching becomes perfect. But the delay is not free, and it is not negotiable:
added_latency_ms = lookahead_msA limiter with 3 ms of lookahead adds exactly 3 ms of delay to everything passing through it, peaks or not. A lookahead limiter reads your audio's mail before delivering it.
In a mastering session nobody cares about 3 ms. In a live monitoring path, where you are hearing your own voice in real time, every millisecond of processing delay stacks on top of your buffer latency, and delays add up fast across a chain of nodes. So Patchd's limiter makes the opposite choice, deliberately: zero lookahead, zero added latency. It accepts letting the first few milliseconds of a sudden peak through, a dB or more over the ceiling, in exchange for being completely safe to leave in a path you are monitoring live. The 0.5 ms attack keeps that window short, though not zero:
samples = attack_ms * fs / 1000 = 0.5 * 48000 / 1000 = 24At 48 kHz the 0.5 ms attack is a 24-sample time constant, not a deadline: the gain closes about 63 percent of the gap to the ceiling in those 24 samples. It only moves while a sample is over the line, though, so on a voice it takes a few cycles to catch a sudden peak, and those first cycles get through a dB or more over the ceiling.
Neither design is wrong. Lookahead is the right call when the audio is not live; zero latency is the right call when it is. What matters is knowing which limiter you are holding, because a mastering-style limiter quietly inserted into a monitoring chain adds delay you will feel and never see labeled.
A ceiling on every channel
In Patchd the limiter is a free node, like the gate and compressor. Add it from the + Add Node menu as the last dynamics stage on a channel, leave the ceiling at -1 dB, and it will sit silently until the one moment you need it, then hold the line with no added delay. The engine itself runs about 10.7 ms through the bus at a 512 buffer, and smaller ASIO buffers take it lower; a zero-latency limiter keeps that number honest. You can try it yourself: Patchd is in open beta now, and if you clear the terms your invitation arrives straight away.