What is LUFS?

LUFS in plain English

Put two songs on a playlist with the same peak level and one can still sound much louder than the other. Peak level only tells you how high the tallest spike in the waveform reaches; it says nothing about how much energy sits underneath it. A ballad with one loud snare hit and a heavily limited dance track can both peak at −1 dBFS, and the dance track will sound several decibels louder all the way through.

LUFS stands for loudness units relative to full scale, and it measures that second thing: how loud the sound comes across to a person. It's defined by the International Telecommunication Union in a recommendation called ITU-R BS.1770, written for broadcasters who needed programmes and adverts to play at a consistent level. Music streaming adopted the same measurement later, which is why LUFS now turns up on mastering plug-ins and in Spotify's advice to artists.

The numbers work like decibels. 0 LUFS is the top of the scale, real music reads as a negative number, and one loudness unit (LU) is the same size as one decibel. A track at −8 LUFS is 6 LU louder than one at −14 LUFS, which is the gap you'd get by turning the quieter one up 6 dB.

Reading the lab

The big screen shows the loudness of the loop over time. The thin pink line is momentary loudness, the thick one is short-term loudness, and the dashed line is the integrated loudness of the whole loop. The pale line behind them is the original loop, so you can always see what your changes did. All of it is measured with the BS.1770 maths on the exact audio you hear.

The meters next to it are the ones you'll find on any loudness meter plug-in: M for momentary, S for short-term, I for integrated and TP for true peak. Below them is the waveform of the whole loop, and further down a screen comparing the crest factor of the original and your master, which is the gap between their peaks and their average level.

You have two controls. Gain is plain volume. The limiter turns the loop up into a ceiling at −1 dBTP, so the peaks stay put and everything under them gets louder, which is how masters are made loud. The buttons above them apply each streaming service's normalisation, and each one shows how far that service would turn your master up or down. A/B switches between your master and the untouched loop, both through the same normalisation.

The Right now panel explains what your settings are doing, in numbers measured from the loop. You can also load one of your own tracks with the Your sound button, and the file stays on your device. The lab loops the first 30 seconds and sets them to the same peak level as its other sounds, so its readings won't match your file's own loudness. Use it to hear what a limiter and each service would do to your mix, and measure the real file with a meter plug-in.

Peak vs RMS vs LUFS

Peak, RMS and LUFS meters all put a number on the level of a sound, but they answer different questions.

A peak meter shows the highest point the waveform reaches. It reacts instantly and tells you how close you are to the top of the digital scale, so it's the meter that warns you about clipping. It's poor at telling you how loud something sounds, because a 2 ms drum transient can set the peak without adding much to the loudness.

An RMS meter averages the level over a short window, usually a few hundred milliseconds. That tracks loudness far better. A sine wave peaking at 0 dBFS reads −3 dB RMS, and a dense, limited mix might sit 8 dB under its peaks while a quiet acoustic recording sits 18 dB under. The weakness of RMS is that it treats every frequency the same, and your ears don't.

LUFS starts from the same idea as RMS and adds what it takes to match hearing. It filters the sound first, so deep bass counts for less and everything above about 2 kHz for a little more, then adds the channels together. For the integrated figure it also ignores silence and very quiet passages. That's why two meters can disagree: a bass-heavy mix can read the same RMS as a bright one and still measure quieter in LUFS, because the filter counts its low end for less.

On a master you'll want a true peak meter to keep you out of clipping and a LUFS meter to tell you how loud the track is next to others. The gap between the two readings shows how much room the dynamics have. The LUFS vs RMS vs peak comparison goes through more examples.

How LUFS is measured

Every LUFS meter, whether it's the free Youlean Loudness Meter, Logic's Loudness Meter or the one in this lab, follows the same recipe from BS.1770, so on the same audio their readings should agree to within about a tenth of a unit.

K-weighting

First the meter runs the audio through a filter called K-weighting, which has two parts. A high shelf lifts everything above about 2 kHz by a few decibels, standing in for the way your head and ears boost those frequencies. A high-pass filter then rolls off the low end below about 100 Hz, more steeply the lower you go, because your hearing is much less sensitive down there at normal listening levels. The figure draws the curve from the same coefficients the lab uses.

Slide it to 40 Hz and you'll see the meter counts it about 6 dB lower than 1 kHz. That's how a kick drum or a sub-bass can push a peak meter hard while adding less to the LUFS reading than you'd expect, and how a bright synth lead at a modest peak level can add a lot of measured loudness.

Averaging the power

After the filter, the meter squares each sample and averages the result over a window, which gives you the signal's power. It does this for each channel and adds the channels together. Left and right count equally, so a sound playing in both channels measures 3 LU louder than the same sound in only one.

The last step turns that power into a number on the LUFS scale using a fixed offset, chosen so that a 997 Hz sine wave peaking at 0 dBFS in one channel reads −3.01 LUFS. The tests behind this lab check that exact case, along with test signals from the European Broadcasting Union's guide to checking meters.

Gating

For integrated loudness, the meter splits the audio into overlapping 400 ms blocks and throws some of them away before it averages. Blocks quieter than −70 LUFS go first, which removes silence. Then the meter works out the average of what's left and drops every block more than 10 LU below it. A song with a long, quiet intro and a loud chorus ends up measured mostly on the parts you'd call its normal level, without the quiet bits dragging the number down.

Momentary, short-term and integrated loudness

A loudness meter runs the same measurement over windows of different lengths.

Momentary loudness (M) covers the last 400 ms. It jumps with every hit and is handy for spotting a single moment that's much louder than the rest, like a drum fill or a shouted word.

Short-term loudness (S) covers the last 3 seconds. It moves slowly enough to read while the music plays, so it's the one to watch while you balance a chorus against a verse. On the lab's big screen it's the thick pink line.

Integrated loudness (I) is one number for the whole track, gated as described above. It's the figure streaming services read when they decide how far to turn a song up or down, and what people mean when they say a master is "at −9 LUFS". In a DAW you get it by playing the whole song from start to finish with the meter running, or with the meter's offline analysis.

Many meters also show loudness range, or LRA, from EBU Tech 3342. It takes the short-term readings across the track, drops the quietest ones and reports the spread between the quieter and louder parts in LU. An orchestral recording can easily have a range of 15 LU or more, while a heavily limited pop master can be under 5. In the lab it's the first stat in the Right now panel.

What is true peak?

A digital file stores the waveform as a series of samples, but the sound coming out of your speakers is a smooth curve drawn through them. That curve can rise above the highest sample, in between two samples, and a sample-peak meter misses it completely. The level the real curve reaches is the true peak, measured in dBTP.

BS.1770 finds true peaks by oversampling: it works out extra points between the samples (four for every original sample at 44.1 or 48 kHz) and looks for the highest one. The figure shows the worst case, a sine at a quarter of the sample rate whose samples all miss the crest, where the samples read −3 dBFS and the wave reaches 0.

On real music the gap is usually a fraction of a decibel. It grows on loud, heavily limited masters, and converting a file to MP3 or AAC can add a little more on top. Spotify recommends keeping the true peak under −1 dBTP, or under −2 dBTP if your master is louder than −14 LUFS. That's why the limiter in the lab has its ceiling at −1 dBTP and listens for true peaks, and why you should switch on the true peak option on your own limiter.

Loudness normalisation on Spotify, YouTube and Apple Music

Streaming services don't want listeners grabbing the volume every time a new song starts, so they use loudness normalisation. Each service measures the integrated loudness of every track and adjusts its playback level towards a target. The file itself isn't changed; the service plays it louder or quieter.

Spotify plays tracks at −14 LUFS by default. Premium listeners can switch to Loud (−11 LUFS) or Quiet (−19 LUFS) in the settings, and normalisation can be switched off there too. Spotify turns quiet masters up as well as loud ones down, but it leaves 1 dB of headroom, so a quiet track with high peaks only goes up as far as those peaks allow.

YouTube aims for about −14 LUFS too, but it only ever turns tracks down. A video that measures −18 LUFS plays at −18.

Apple Music uses a feature called Sound Check and plays tracks at −16 LUFS, turning quieter tracks up towards that level as well as bringing louder ones down.

Say your master measures −8 LUFS integrated. Spotify plays it 6 dB quieter, at −14, and Apple Music 8 dB quieter, at −16. A more dynamic master of the same song at −14 LUFS isn't touched by Spotify at all and comes down just 2 dB on Apple Music, so on both services the two versions end up equally loud. What the listener hears is the difference in shape.

Click through the services in the lab and each button shows the change it would make. Then press A/B to hear the original loop through the same service. The streaming loudness lesson covers album mode, playlists and the settings listeners can change.

Is a louder master better?

On a normalised service, a louder master plays at the same level as a quieter one, with smaller peaks.

A limiter can only make a master louder at the same peak level by turning the loudest moments down, so each drum hit loses some of its height compared with everything around it. The crest factor shrinks. When Spotify then turns the whole master down to −14 LUFS, those reduced peaks come down with it. Next to a dynamic track at the same loudness, the squashed one has less snap on the drums and can sound flatter and smaller, even though it was the louder file.

You can watch it happen in the lab. Push the limiter to 14 dB and switch on Spotify, and the crest factor screen shows the master's peaks several decibels below the original's once both are playing at −14 LUFS.

Plenty of electronic, hip-hop and rock music is still mastered loud on purpose, because the dense, pressed sound is part of the style, and DJs often play files in software that doesn't normalise. If the loud version sounds better to you at matched loudness, it's the right master. Just make the comparison at matched loudness, which is what normalisation and the lab's A/B both do for you.

How loud should you master?

There isn't one right number, but there are sensible ranges. The table lists the published targets for the main places music and audio end up.

For music, a master between −14 and −10 LUFS with the true peak under −1 dBTP plays well everywhere. At −14 nothing gets turned down on Spotify or YouTube, and towards −10 you get a denser sound that still keeps most of its punch. Going much louder than that mostly trades dynamics for level the services take away again. If you're releasing for DJs, or in a genre that's always mastered hot, measure a few reference tracks from that genre with the same meter and decide from there.

For broadcast, or anything else with a delivery spec, the number in the spec is the rule. EBU R128 deliveries have to land on −23 LUFS within a small tolerance, and broadcasters check files against it.

How to master to a LUFS target, step by step

1. Put a loudness meter last on the masterAfter the limiter, so it reads exactly what you'll export. Youlean Loudness Meter has a free version, and Logic comes with one called Loudness Meter.
2. Set the limiter's ceiling to −1 dBTPSwitch on true peak detection if your limiter has it. If it doesn't, set the ceiling a little lower, around −1.5 dB, and check the true peak reading on the meter.
3. Play the loudest section and raise the limiter's inputWatch the short-term reading while the chorus plays. Stop when it sounds as dense as you want, or as soon as the limiter starts to dull the drums.
4. Measure the whole songReset the meter and play from the very start to the end, or use its offline analysis. The integrated figure is the one that counts.
5. Compare at the streaming levelTurn your master down to −14 LUFS with plain gain and listen next to a reference track that's also at −14. If you prefer the sound with less limiting, back the input off and measure again.
6. Check the exported fileLoad the exported file into the meter. Its integrated loudness and true peak should match what you measured in the session.

Common LUFS mistakes

Measuring one loud section

A chorus played on its own reads louder than the whole song, because the integrated figure includes the verses and the breakdown. Read the short-term meter while you work on a section, and measure the integrated loudness across the whole track before you compare it with a target.

Treating −14 as a goal

−14 LUFS is where Spotify plays your track. A quiet piano piece mastered at −18 LUFS sounds fine on Spotify and gets turned up a little if its peaks allow; pushing it to −14 with a limiter to hit the number just squashes it.

Ignoring true peak

A limiter set to a −0.1 dB sample-peak ceiling can let true peaks through above 0 dBTP, and those can distort when a service encodes the file. Use a true peak ceiling and keep it at −1 dBTP or lower.

Judging loudness with a peak meter

Two masters that both peak at −1 dBTP can be 6 LU apart in loudness. When you match your track to a reference, match them in LUFS. The quiz under the lab is built on this, and in plenty of its rounds the louder version has the lower peaks.

Comparing with a reference at a different level

Louder nearly always sounds better for the first few seconds, so a reference that's 3 LU louder than your mix makes your mix sound worse than it is. Turn whichever one is louder down until both measure the same before you compare tone or punch.

Training your ears to hear loudness

Ears are good at picking the louder of two sounds when the gap is large. At 1 or 2 LU they're easy to fool, especially when the quieter one has bigger peaks. The quiz under the lab plays two versions of the loop with the screens covered and normalisation off, and asks which one has the higher integrated loudness. The louder one is sometimes the limited version and sometimes the untouched loop turned up, so listening for squashing won't help. Flip between A and B a few times before you choose and listen to the overall level of the loop, more than to single hits.

Questions people ask

Keep going