Peak vs RMS compression

At a glance

What does a compressor's detector do?

Every compressor has two paths. The audio goes through a gain stage that can turn it down, and a copy of it goes to the sidechain, where the compressor works out how loud the sound is. The part that does the measuring is the detector. Its reading is what gets compared with the threshold, and the ratio, attack and release all act on whatever it reports.

So the same threshold can mean different things on two compressors. Set −20 dB on one that reads peaks and on one that reads RMS, play them the same drum loop, and they'll disagree about when the drums are over the line and by how much. If you're new to the controls themselves, what is compression? covers them first.

What is peak detection?

A peak detector takes the size of each sample, ignoring whether the wave is above or below the centre line, and passes it straight on. When a snare hit arrives at −1 dBFS, the detector reports −1 dBFS on that very sample, so nothing can get past it unnoticed, which is what a limiter needs.

In the lab with Peak selected, the pink shape on the detector screen is that reading. It jumps up at the start of the zoomed hit and flickers with every cycle of the wave underneath, because a waveform passes through zero twice per cycle. The attack and release smooth those jumps into the gain you hear, so a peak compressor doesn't normally chatter. Give it a very short release, though, and the gain starts to follow the flicker, which is where peak detection gets into trouble on bass.

On the drum loop at a −16 dB threshold, 4:1 and a 0.5 ms attack, the peak detector takes up to 9.9 dB off, and the loudest sample of the zoomed hit is already 7.0 dB down by the time it plays. The highest peak in the loop comes out at −5.8 dBFS instead of −0.9 dBFS.

What is RMS detection?

RMS stands for root mean square, which names the steps the detector takes in reverse order. It squares every sample in a short stretch of time, called the window, takes the mean of those squares, then takes the square root of that. Squaring makes every value positive and weights the bigger ones more, so the result tracks how much energy the sound has over the window.

The window is what makes RMS slow. A spike that lasts 2 ms inside a 30 ms window only fills a small part of the average, so the detector reads it much lower than its real height. It reads it late too, because the average has to fill up first, and it holds on afterwards until the spike slides out of the other end of the window.

For a steady sine wave the RMS level is always 3 dB under the peak, because the square root of the mean of a sine squared is 0.707 of its height. Real sounds have a bigger gap. On the drum loop, a 30 ms RMS detector reads the loudest moment at −8.3 dB, 7.4 dB under its −0.9 dB peak. The gap between peak and average is called the crest factor, and the spikier the sound, the bigger it is.

RMS is popular for compression because it lines up with how loud something feels. Your ears add up loudness over time as well, over roughly the last tenth to a fifth of a second. A sound with lots of energy across that stretch feels loud even if its peaks are modest, while a single sharp click feels small even if its peak is high. A compressor with an RMS detector turns down the parts that sound loud.

Why RMS needs a lower threshold

Flip the lab from Peak to RMS without touching anything else and the compressor seems to back off. On the drums at −16 dB it goes from 9.9 dB of gain reduction to 5.8 dB, and from working on 66% of the loop to 33%. Nothing is broken. The RMS detector reads every hit several decibels lower, so less of the loop gets over the threshold, and what does get over gets over by less.

To compare the two fairly, lower the threshold until the RMS compressor is doing a similar amount of work. At −20 dB it takes up to 8.8 dB off and works on 58% of the loop, close to what peak did at −16 dB. On most material you'll need somewhere between 3 and 10 dB of difference, more on spiky drums and less on a sustained pad.

What happens to transients

With the two compressors taking a similar amount off, look at the gain close-up in the lab. They're turning the drums down at different moments.

The peak compressor is most of the way down within a millisecond of each hit arriving, so the transient, the short spike at the front, gets the most gain reduction of anything in the loop. The RMS compressor is still filling its window when that spike goes past. At −20 dB the loudest sample of the zoomed hit is only 0.2 dB down, and the gain comes down over the next few tens of milliseconds, on the body and ring of the drum. The loop's highest peak comes out at −1.1 dBFS, almost exactly where it went in.

That's why RMS compression on drums tends to sound punchier. The crack is left alone and everything behind it is pulled down, so the hit stands out more against its own tail. Peak compression with a fast attack does the opposite and makes drums sound smaller and further away. You can get some of the RMS sound from a peak compressor by slowing the attack, and fast vs slow attack covers that in detail.

The RMS window

A compressor with an RMS detector has to choose how long to average over, and some let you choose. Ableton Live's Compressor has Peak and RMS buttons, while Reaper's ReaComp has an RMS size control in milliseconds that gives you a peak detector at 0. The lab has a Window knob that goes from 1 ms to 300 ms.

Longer windows read lower, react later and let go later. At 150 ms the drum loop's loudest moment reads −13.3 dB, the compressor takes up to 5.1 dB off at the same −20 dB threshold, and the gain lane drifts with the groove instead of dipping on each hit. Very long windows start to behave like a slow levelling amplifier riding the volume of a whole phrase.

Shorter windows read closer to the peak. At 1 ms there's almost nothing to average, so the reading comes up to −3.2 dB and the compressor has the loudest sample 8.1 dB down, nearly as hard as the peak detector managed. An RMS detector with a tiny window is a peak detector that reads a few decibels low.

For anything with low notes in it, keep the window at least as long as one cycle of the lowest note. A cycle at 55 Hz lasts 18 ms, and a window shorter than that rises and falls with the wave itself, which brings back the distortion problem in the next section.

Peak vs RMS on bass

Bass is where the two detectors sound most different, and it comes down to the release. A peak detector's reading falls to nothing twice in every cycle of the wave. A long release ignores that. A release shorter than the cycle doesn't, so the gain rises between the tops of the wave and comes back down on each one, which reshapes the wave and adds harmonics. You hear it as a buzzy, gritty edge on the low notes.

With the bass loop through a peak compressor at −18 dB, 4:1 and a 10 ms release, the gain moves by 2.0 dB within every 10 ms on average while it's working. Switch to RMS with a 30 ms window and lower the threshold to −22 dB so it takes a similar amount off, and that wobble drops to 0.5 dB. The window spans more than a full cycle of the notes, so the reading stays almost still from one cycle to the next.

You can also fix the peak version by lengthening the release past 50 ms or so, and plenty of people compress bass with peak detection that way. RMS makes the problem harder to cause in the first place.

Which compressors use peak and which use RMS

Most hardware compressors don't have a switch. The detector is built into the circuit, and it often sits somewhere between pure peak and pure RMS. The dbx 160 was sold on its RMS detector, which dbx called true RMS. The Teletronix LA-2A turns the sound down with a light and a light-sensitive cell that takes time to respond, so it behaves more like an averaging detector than a peak one. Fast FET designs like the Universal Audio 1176 react quickly enough to catch the front of most transients.

In plug-ins, a peak and RMS switch is common on stock compressors, and plug-ins modelled on hardware usually copy the original's detector without telling you which kind it is. Limiters almost always use peak detection, and good ones read true peak, the level between the samples.

The same split exists in meters. A peak meter shows the highest sample, an RMS meter shows the average and sits lower, and loudness meters in LUFS go a step further. LUFS vs RMS vs peak compares all three.

How to choose between peak and RMS

1. Decide what the compressor is forIf it's there to stop hits jumping out, pick peak. If it's there to make the part sound more even from line to line or bar to bar, pick RMS.
2. Set the threshold for the detector you chosePeak thresholds sit nearer the tops of the hits. RMS thresholds sit 3 to 10 dB lower for the same amount of compression, so set them by watching the gain reduction meter instead of copying a number.
3. With RMS, set the window if you canStart at 20 to 50 ms. Make it longer for smoother control and shorter if it's letting through more than you want.
4. Listen to the front of the hitsIf drums have lost their snap under peak detection, slow the attack or try RMS. If they poke out too far under RMS, shorten the window or switch to peak.
5. Check the low endOn bass, listen for grit. Lengthen the release, or switch to RMS with a window longer than one cycle of the lowest note.
6. Compare at the same loudnessUse A/B with matched levels before deciding. The lab does this for you.

Which detector for which sound

These are starting points, chosen for what the compressor is usually doing on each source.

Common mistakes

Keeping the threshold when you switch

Going from peak to RMS at the same threshold makes the compressor do far less, and going the other way makes it do far more. Reset the threshold by watching the gain reduction meter every time you change the detector.

Expecting RMS to stop peaks

An RMS compressor lets short peaks through by design, so it won't protect you from overs. In the lab at −20 dB, the drum loop still peaks at −1.1 dBFS while up to 8.8 dB of gain reduction is going on. If the peaks need controlling as well, put a peak compressor or a limiter after it.

A short release with a peak detector on bass

It's the quickest way to get distortion out of a compressor that isn't meant to distort. If the low end gets fuzzy as you compress, lengthen the release or switch to RMS.

Treating the detector switch as a tone control

Peak and RMS change when the compressor reacts and how much. Any change in tone comes from that, like the grit on bass with a short release, so if one sounds brighter or fuller in a quick comparison, check that it isn't simply louder or compressing less before you put it down to the detector.

Training your ears to hear the difference

The quiz under the lab plays a sound through one of the two detectors with the controls covered. Each round sets the threshold so both take up to 8 dB off, and the levels are matched, so the only clue is how the compressor moves. On drums, listen to the very front of each hit: a softened front means peak, and a sharp front with a quieter tail means RMS. Start on the drums, because their hits have the biggest gap between peak and average, and the bass and vocal are harder.

Questions people ask

Keep going