What is loudness normalisation?

Loudness normalisation in plain English

Without normalisation, a song plays at whatever level it was mastered at. Put a quiet 1970s recording and a modern pop single on the same CD player and you’d be reaching for the volume knob between them. Streaming services fixed that by doing the reaching for you.

When a track is uploaded, the service measures its integrated loudness, a single number for how loud the whole song sounds to a listener. It compares that number with its own target and works out the difference. When the track plays, the app turns the volume up or down by exactly that difference, so a song that measures −8 LUFS on Spotify is played 6 dB quieter, at −14.

That one gain change applies to the whole song. It doesn’t touch the balance, the EQ or the dynamics, and the file you uploaded stays exactly as it was on the service’s servers. Normalisation here is the same job as moving a fader before the song starts and leaving it there.

This page goes through what each service does and where they differ, then what that means for how loud you master. If you want to know how the LUFS number itself is measured, start with What is LUFS?, which covers K-weighting, gating and the meters.

Reading the lab

The big screen shows the loudness of the loop over time, with your master in pink and the untouched loop as a pale line. The dashed pink line is your master’s integrated loudness, the number a streaming service reads. Pick a service from the row of buttons under the screens and a teal line appears at its target, and every line is redrawn at the level that service would play it.

Each service button also says what it would do to your master before you press it, such as “down 4.1 dB” or “no change”. Change the master with the gain and limiter knobs and those figures move with it.

The four meters beside the big screen read what you’re hearing, including the service’s change. The crest factor screen compares the peaks and average level of the original and your master, and A/B switches between them, always through the same service. The panel marked Right now describes what’s happening in numbers measured from the loop. When the lab turns a track up for Apple Music it stops at −1 dBTP, the same as Spotify; Apple doesn’t publish its exact rule for peaks.

Your own track works too. The Your sound button loads a file that stays on your device, and the lab loops its first 30 seconds at the same peak level as its other sounds, so for real numbers measure the whole exported file with a loudness meter.

Normalisation on Spotify, YouTube and Apple Music

Spotify and Apple Music measure loudness with the ITU-R BS.1770 method behind LUFS. YouTube doesn’t publish its method. The services differ on the target, on whether they’ll turn quiet tracks up, and on what listeners can change.

Spotify

Spotify plays tracks at −14 LUFS by default, which it calls Normal. Premium listeners can pick Loud, at −11 LUFS, or Quiet, at −19 LUFS, from the volume level setting, and normalisation can be switched off in the same part of the settings.

Spotify turns quiet tracks up as well as loud ones down. When it turns a track up it leaves 1 dB of headroom, so it stops if the peaks would come within 1 dB of full scale. The Loud setting is the exception: there Spotify puts a limiter on the playback so it can bring quiet tracks up to −11 LUFS without them clipping.

Its advice to artists is to keep the true peak under −1 dBTP, or under −2 dBTP for masters louder than −14 LUFS, because louder masters tend to gain more peak level when they’re encoded to the compressed formats Spotify streams.

YouTube

YouTube only turns tracks down. It doesn’t publish its target, but the level people work to is about −14 LUFS, and a video quieter than that plays at whatever level it was uploaded. There’s no setting on the website to switch it off.

You can see what YouTube is doing to any video. Right-click the player, choose Stats for nerds, and look at the Volume / Normalized line. A reading like “100% / 63% (content loudness 4.0dB)” means the video sits 4 dB above YouTube’s reference and is being played at about 63% volume to make up for it. A negative content loudness means the video is quieter than the reference and isn’t being changed.

Apple Music

Apple calls its normalisation Sound Check. With it on, tracks play at −16 LUFS, a couple of decibels lower than Spotify, and quieter tracks are brought up towards that level as well as louder ones brought down. Listeners can switch Sound Check off in the Music settings.

Because the target is lower, Apple Music turns a typical loud master down further than Spotify does. A −9 LUFS master comes down 5 dB on Spotify and 7 dB on Apple Music.

Other services

Tidal and Amazon Music also have a loudness normalisation setting, and their targets sit in the same region. They change their settings from time to time, so check their own help pages if one of them is your main outlet. Downloads, CDs and DJ software are a different case: unless the player says it normalises, assume it plays your file at the level you exported.

Loud masters get turned down

Say your master measures −8 LUFS integrated, with a true peak of −1 dBTP. Spotify plays it 6 dB quieter, at −14 LUFS, so its peaks now only reach −7 dBFS when it’s played. On Apple Music it’s 8 dB quieter, with peaks at −9.

Now take a gentler master of the same song at −12 LUFS, also peaking at −1 dBTP. Spotify plays it 2 dB quieter, also at −14 LUFS, with its peaks at −3 dBFS. Both versions are now equally loud, and the gentler one has 4 dB more room above its average level for drum hits and accents. The extra loudness the hard limiting bought has been taken away, and what’s left is the cost: smaller peaks and a flatter sound.

The lab shows the same thing with the full mix. With 10 dB of limiter push the master reads about −10 LUFS, Spotify takes about 4 dB off, and its peaks end up near −5 dBFS. The untouched loop, which already measures about −14 LUFS, plays at the same loudness with its peaks reaching about −0.2 dBFS. On the crest factor screen the gap shows up as two columns of the same height with very different peak lines.

A loud master can still be the better one. Plenty of electronic music and rock is mastered loud because the dense sound is part of the style, and some tracks hold up well at −8 LUFS. Loudness on its own stops being an advantage, though, so the only reason left to limit hard is that you prefer the sound of it at the same playback level.

Will a quiet master get turned up?

On YouTube, no. On Spotify and Apple Music it depends on the master’s peaks, because turning a track up raises its peaks by the same amount.

Suppose your master measures −20 LUFS with its true peak at −6 dBTP. Spotify wants to add 6 dB to reach −14 LUFS, but 6 dB would take the peaks to 0 dBTP. It only goes as far as −1 dBTP, which is 5 dB, so the track plays at −15 LUFS, a decibel under everything else. The lab’s quiet master is almost exactly this case: Spotify adds 5.2 dB and it plays at −14.9 LUFS.

The amount of turning up a track can take is the gap between its true peak and −1 dBTP. A master with a large crest factor, the gap between its peaks and its average level, runs out of that headroom first. Solo piano and orchestral recordings often measure −18 to −23 LUFS with peaks close to the top, so on Spotify they play a little quieter than the pop around them, and on YouTube a lot quieter.

Usually that’s fine. A quiet ballad is meant to be quieter than the single before it, and a listener will turn it up if they want. If you’d prefer it to sit level with other tracks, a few decibels of gentle limiting raises the loudness while it lowers the peaks, which gives the service room to bring it up. In the lab, 3 dB of limiter push on the quiet master is enough for Spotify to bring it all the way to −14 LUFS.

Album normalisation and track normalisation

An album is often mastered with a shape in mind: a quiet intro, a loud single, a softer closing track. If every song were set to −14 LUFS on its own, that shape would be flattened, and the intro would come up to the same level as the single.

Spotify avoids this with album normalisation. When someone plays an album from start to finish, Spotify turns every track on it by the same amount, so the differences between songs stay as you mastered them. When a song plays in a playlist or on shuffle, it’s normalised on its own, which is track normalisation, so it fits in next to music from other records.

Here’s a worked case. Say an album has a single at −10 LUFS and an interlude at −16 LUFS, 6 LU quieter. Played as an album, both get the same change and the interlude stays 6 LU below the single. In a playlist the single comes down 4 dB to −14 and the interlude goes up as far as its peaks allow, so the 6 LU gap shrinks or disappears. Master the album so the quiet tracks sound right next to the loud ones, and let playlists handle the rest.

How loud should you master for streaming?

Loud enough to sound the way you want, with the true peak under −1 dBTP. For most music that lands somewhere between −14 and −9 LUFS, and you don’t need to hit −14 exactly.

Below −14 LUFS you lose nothing on Spotify as long as the peaks leave room to be turned up, though you will play quieter on YouTube. Between −14 and about −10 you get a denser sound that still keeps most of its punch, and the services take off a few decibels. Past −9 or so, most masters are trading their transients for loudness that every normalising service removes again. The normalising vs limiting comparison goes further into why those two get confused.

Keep the true peak at −1 dBTP or lower whatever loudness you choose, and lower still for very loud masters. Encoding to Ogg Vorbis, AAC or Opus can add a little peak level, and a peak that passes 0 dBTP after encoding can distort on playback. The true peak lesson shows where those peaks come from.

If you’re releasing mostly for DJs or on vinyl, normalisation doesn’t apply there, so measure a few reference tracks from your genre with the same meter and match their range.

How to check what a service will do to your master

1. Measure the exported fileLoad the final WAV into a loudness meter, such as the free version of Youlean Loudness Meter, and analyse the whole song. Write down the integrated loudness and the true peak.
2. Work out the change for each serviceSubtract your loudness from the target. A −9.5 LUFS master gets −4.5 dB on Spotify and −6.5 dB on Apple Music; YouTube takes off about 4.5 dB as well.
3. If the change is upwards, check the peaksA track can only go up by the gap between its true peak and −1 dBTP on Spotify. If that gap is smaller than the change you worked out, the track will play under the target, and on YouTube it won’t move at all.
4. Listen at the playback levelTurn your master down with plain gain until it measures −14 LUFS and play it next to a reference track also set to −14. This is how most listeners will hear the two side by side.
5. Try a gentler versionBack the limiter off by 2 or 3 dB, set it to −14 LUFS the same way and compare again. If you can’t hear it getting worse, the gentler master is the better one to release.
6. Check it once it’s liveOn YouTube, the Stats for nerds panel shows the content loudness of your upload. It should be close to the figure you worked out in step 2.

Common normalisation mistakes

Thinking normalisation compresses your track

It doesn’t. On Spotify’s Normal setting and on YouTube a loud track is simply played quieter, with every part of it moved by the same amount. If your song sounds flatter than another on Spotify, the flattening happened in your limiter before you uploaded it.

Limiting a quiet song to hit −14

A dynamic acoustic track at −18 LUFS is meant to be quieter than a pop single. Squashing it to −14 to match a number makes it louder only on the services that wouldn’t have turned it up anyway, and costs the dynamics everywhere.

Comparing with the normalisation switched off

If your Spotify app has normalisation off, your loud master really does play louder than a gentler reference, and it will seem to win. Check the setting before you judge your track against others in the app.

Measuring a section instead of the song

Services measure the whole track. A chorus measured on its own reads a few LU louder than the full song, so a master you think is at −14 LUFS may actually be at −16 and get turned up, or, on YouTube, play quieter than you expected.

Letting the true peak sit at 0 dBTP

A master with a true peak at −0.1 dBTP has no room to be turned up on Spotify at all, and its peaks can go over full scale once it’s encoded. Set your limiter’s ceiling to −1 dBTP with true peak detection on.

Training your ears for normalised playback

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. One is often the limited master and the other the untouched loop turned up, so the version with the bigger peaks is sometimes the quieter one. A streaming service makes the same judgement with a meter when it sets your track’s level.

Questions people ask

Keep going