LUFS vs RMS vs peak

At a glance

Same peak, different loudness

The quickest way to see why there are three meters is to put two very different sounds side by side. In the lab, the acoustic guitar loop and the distorted drums loop both have a sample peak of −0.9 dBFS, and their true peaks are a tenth of a decibel apart. On a peak meter they look the same.

They don't sound the same. The guitar is plucked, so each note starts with a sharp spike and then rings away into near silence, and most of the loop sits far below its peaks. The distorted drums have been squashed and saturated until there's very little space between the hits. Their RMS levels are 9 dB apart (−25.7 against −16.7 dB), and their loudness is 8.2 LU apart (−21.8 against −13.6 LUFS). Played one after the other at those levels, the drums sound far louder.

Turn the drums down 8 dB and they measure −21.6 LUFS, within 0.2 LU of the guitar, so now they play at about the same loudness. Their peaks are 8 dB lower than the guitar's. Matching peaks gave you two very different loudnesses, and matching loudness gives you two very different peaks. The gap between peak and average is the crest factor.

What is a peak meter?

A peak meter shows the highest point the waveform reaches, in dBFS. 0 dBFS is the largest number a fixed-point audio file can store, so the peak meter is the one that tells you how much headroom you have left and whether the file will clip when you export it. The meters on your DAW's channels are usually peak meters, often with a hold marker that keeps the highest recent reading on screen.

A peak meter reacts instantly, which is what makes it good at catching overs and useless for loudness. A 2 ms snare transient can set the peak for a whole song while adding almost nothing to how loud the song sounds.

Sample peak and true peak

A basic peak meter reports the largest sample. The audio that reaches your speakers is a smooth curve drawn through those samples, and the curve can rise higher than any of them, in between two samples. A true peak meter finds that higher point by oversampling: at 44.1 or 48 kHz it rebuilds the wave at four times the sample rate, three new points between every pair of samples, and reports the highest one in dBTP.

The bass loop in the lab shows the difference. Its highest sample is −0.5 dBFS, comfortably under full scale, but its true peak is +0.03 dBTP. A sample-peak meter would call that safe, while a converter or an MP3 encoder might clip it. The true peak lesson covers the details; on the master, use true peak.

Broadcast peak programme meters (PPMs) are slower than either. They're quasi-peak meters that take 5 to 10 ms to reach a reading, so they show the shortest transients lower than a sample-peak meter does. You won't need one for music, but it explains why an old broadcast desk and a plug-in can disagree on the same peak.

What is an RMS meter?

RMS stands for root mean square, which describes the maths: square every sample, take the mean of those squares over a stretch of time, then take the square root. Squaring makes every sample positive and gives big values more weight, so the result tracks the power in the signal. Power is much closer to how loud something sounds than the single highest sample is.

A sine wave is the easy worked example. Its RMS level is always 3.01 dB under its peak, so the lab's 110 Hz sine, peaking at −0.9 dBFS, reads −3.9 dB RMS. Music has a much bigger gap because it's full of short peaks with less going on between them. The full mix in the lab sits 16.2 dB under its peak, and the acoustic guitar 24.8 dB.

Windows and the 3 dB question

An RMS meter has to choose how long a stretch to average over. Around 300 ms is common, which is about the response time of a VU meter on a mixing desk, a close cousin of RMS. A shorter window jumps around with every hit, and a longer one moves slowly and hides short loud moments. Two plug-ins with different windows will show different RMS readings on the same audio, and neither is wrong.

RMS meters also disagree on where zero is. Measured straight, as this lab does it, a sine peaking at 0 dBFS reads −3 dB RMS, while the AES17 standard adds 3 dB so that the same sine reads 0 dB, and many meter plug-ins let you switch to the AES17 reading. If your RMS readings look 3 dB higher than you expect, check which one your meter uses.

Silence drags RMS down

Plain RMS counts every moment equally, silence included. Take the lab's 8-second drum loop and add 8 seconds of silence after it: the RMS of the file drops by 3 dB, from −18.4 to −21.4 dB, even though the drums are just as loud as before. A track with a long, quiet intro or big gaps between phrases reads quieter on an RMS meter than it sounds while it's playing.

Bob Katz's K-System, a metering scheme from the years before LUFS, was built around RMS-style average meters with the zero mark set 12, 14 or 20 dB below full scale. You'll still see K-14 and K-20 scales on some meter plug-ins.

What LUFS adds to RMS

LUFS starts from the same mean square that RMS uses and changes it in the ways ITU-R BS.1770 sets out, so every LUFS meter gives the same answer on the same audio. The LUFS lesson goes through each stage and the time windows in detail; here the question is what each change does to the reading next to RMS.

K-weighting filters the sound

Before averaging, a LUFS meter runs the audio through K-weighting. A high-pass filter rolls off the low end, more steeply the lower you go, and a shelf lifts everything above about 2 kHz by a few decibels. The figure draws the curve from the same coefficients the lab's meter uses.

At 40 Hz the meter counts the sound 6.3 dB lower than at 1 kHz, and at 110 Hz it's 1.6 dB lower. An RMS meter counts them all the same. So a mix with a big sub-bass can read high on an RMS meter while sounding no louder than a brighter mix, and LUFS gets that comparison closer to what you hear.

The channels are added together

An RMS meter shows each channel separately, or averages them. A LUFS meter adds the channels' energy together, so a sound playing in both speakers reads 3 LU louder than the same sound in one. That's most of why LUFS readings on a stereo mix come out higher than RMS readings.

Put the two together for the 110 Hz sine: −3.9 dB RMS, plus 3 dB for the two channels, minus the 1.6 dB K-weighting takes off at 110 Hz, gives the −2.5 LUFS the lab reads. The standard also has a fixed offset of −0.691 dB, which cancels the small lift K-weighting gives at 1 kHz, so a 1 kHz tone needs no correction.

Gating leaves silence out

For the integrated figure, a LUFS meter splits the track into 400 ms blocks, throws away any block quieter than −70 LUFS, then also drops any block more than 10 LU below the average of what's left. The drum loop with 8 seconds of silence added still measures −15.8 LUFS integrated, exactly what it measured without the silence, while its RMS fell by 3 dB.

The lab's loops on all three meters

Here are the loops in the lab as they load, measured with the same code the meters use. They're all set to about the same peak level, which makes the other columns easy to compare.

Read down the peak columns and the loops look almost the same. The RMS and LUFS columns spread across about 20 dB, and that spread is how differently loud they really are. The last column is where RMS and LUFS part ways: the sine, which is all low-frequency energy, gets the smallest boost, and the acoustic guitar, with its bright pick attack, the biggest. Work it through for a 40 Hz sine and LUFS would come out about 3 dB below the RMS reading.

The full mix and the bass loop make a useful pair. Their RMS levels are only 0.5 dB apart, and their LUFS readings are 0.5 LU apart too, because there's enough midrange in this bass part for the filter to count. On a mix that's mostly 40 Hz sub, the two meters would disagree by several decibels.

What a limiter does to each reading

Gain is the simple case. Turn a track up 3 dB and the peak, the RMS and the LUFS readings all go up 3 dB, so the gaps between them stay exactly where they were.

A limiter is where the meters split. It holds the peaks at a ceiling and lets everything under them come up, so the peak meter stays put while both averages rise. Push the full mix 8 dB into the lab's limiter, with its ceiling at −1 dBTP, and the true peak sits just under −1 dBTP, the RMS level goes from −16.4 to −12.6 dB and the integrated loudness from −14.1 to −10.3 LUFS. The crest factor shrinks from 16.2 to 11.6 dB.

Push it to 14 dB and the loudness only reaches −9.4 LUFS, less than 1 LU more for the extra 6 dB of push, because by then the limiter is turning down most of the loop as well as the peaks. You can only see that diminishing return on an average meter. On a peak meter, 4 dB of push and 14 dB of push look the same.

Which meter streaming services read

Spotify, YouTube and Apple Music all measure integrated LUFS and adjust the playback level towards their target (−14 LUFS on Spotify and YouTube, −16 on Apple Music). None of them looks at RMS, and peak only comes into it as a limit: Spotify won't turn a quiet track up so far that its true peak passes −1 dBTP.

The limited full mix from the last section measures −10.3 LUFS, so Spotify plays it 3.7 dB quieter. The untouched loop, at −14.1 LUFS with its peaks already close to 0 dBFS, stays where it is because turning it up would push those peaks past −1 dBTP. Both then play at about −14 LUFS, and the original's peaks reach 4.6 dB higher than the master's. The streaming loudness lesson covers each service's settings.

Watch out for the word normalise in your DAW. The Normalize command in most DAWs and audio editors is peak normalisation by default: it turns a clip up until its highest sample hits a level you choose, usually 0 dBFS or a little under, and pays no attention to loudness. Two clips peak-normalised to −1 dBFS can still be 8 LU apart, as the guitar and drums show. The normalising vs limiting comparison goes further.

Which meter to use when

Compressors are the one place RMS is still everywhere. Many let you choose whether their detector reacts to peaks or to an RMS average, and that changes how they sound: peak detection grabs transients, and RMS detection follows the body of the sound more smoothly. The peak vs RMS compression lesson has a lab for it.

How to set up metering on your master

1. Put a loudness meter lastPlace it after the limiter so it reads exactly what you'll export. Youlean Loudness Meter has a free version that shows integrated, short-term and momentary LUFS along with true peak.
2. Switch on true peakSet the meter, and your limiter if it has the option, to true peak. FabFilter Pro-L 2 and most current mastering limiters have it.
3. Watch short-term LUFS while you workIt moves slowly enough to read during playback and shows how sections compare.
4. Measure the whole song for the integrated figureReset the meter and play from start to finish, or use the meter's offline analysis.
5. Read the gapTrue peak minus short-term loudness gives you a rough sense of how much room the dynamics have. If it keeps shrinking as you push the limiter, you're trading punch for level.

Common mistakes

Judging loudness by peak

Matching your mix to a reference by making both peak at −1 dBFS can leave them several LU apart, and the louder one will nearly always sound better at first. Match them in LUFS.

Comparing RMS readings from different meters

One meter might use a 300 ms window and the AES17 convention, another a 50 ms window without it, and the same audio will read a few decibels apart. If you use RMS at all, use one meter for everything.

Reading LUFS as a clipping warning

A track can measure a modest −14 LUFS and still have true peaks over 0 dBTP, and the bass loop, at −14.6 LUFS with a true peak of +0.03 dBTP, is one. Loudness and peak are separate readings, and you need both on the master.

Measuring a loop or a chorus and calling it the song

Integrated loudness covers the whole track, verses and breakdowns included, so a chorus measured on its own reads louder than the song will. Use short-term for sections and integrated for the finished file.

Training your ears for loudness over peaks

The quiz under the lab plays two versions of the same loop with the screens covered and asks which has the higher integrated loudness. A is your master, often pushed into the limiter, and B is the untouched original, each set to its own loudness, so the louder version sometimes has the lower peaks. After each answer the lab tells you when a peak meter would have picked wrong. Listen to the level of the whole loop, since the single hits are what a peak meter sees.

Questions people ask

Keep going