Normalise when a file is simply too quiet or too loud and its peaks have room to move. It’s a volume change, so it can’t make a track any louder than its single highest peak allows.
Limit when the peaks are already near the top and you still want more loudness. The limiter flattens those peaks so the average level can rise, which changes the sound a little, and more the harder you push.
At a glance
| Property | Peak normalising | Loudness normalising | Limiting |
|---|---|---|---|
| What it does | Moves the whole file until its highest peak hits a target | Moves the whole file until its integrated loudness hits a target | Turns the loudest moments down so everything else can come up |
| The gain | One fixed amount | One fixed amount | Changes from moment to moment |
| What it reads | Sample peak, or true peak if you pick it | Integrated LUFS | Each peak as it arrives, with lookahead |
| Crest factor | Unchanged | Unchanged | Smaller |
| Loudness at the same peak | Unchanged | Unchanged | Higher |
| Can it clip? | Not with a target under 0 dBFS | Yes, if turning up pushes peaks over | No, the peaks stay under its ceiling |
| Where you meet it | The Normalize command in a DAW or audio editor | Streaming services, broadcast, some editors | A plug-in, usually last on the master |
| Can you undo it? | Yes, turn it back down | Yes, turn it back | No, flattened peaks stay flat |
What is normalising?
Normalising is a volume change worked out for you. The software measures the file, compares the result with a target you set, and turns the whole file up or down by the difference. Every sample moves by the same number of decibels, so it’s exactly what you’d get by moving the clip’s gain or a fader yourself, without the guesswork.
Because the gain never changes during the file, the shape of the audio is untouched. The loud parts and the quiet parts move together, so the gap between them, its dynamic range, stays the same. So does any hiss or hum: if you turn a recording up 10 dB, the noise under it comes up 10 dB too.
Peak normalisation
This is what the Normalize command in most DAWs and audio editors does unless you tell it otherwise. It finds the single highest peak in the file and sets the gain so that peak lands on your target, often 0 dBFS or a little under. Say a vocal take peaks at −12 dBFS and you normalise it to −1 dBFS: the editor adds 11 dB to every sample, and the vocal’s loudness reading goes up by 11 too.
In the lab, the first step plays the full mix exported 9 dB too low, reading −23.1 LUFS with a true peak of −9.2 dBTP. Normalising it to −1 dBTP means adding 8.2 dB, and the mix lands at −14.9 LUFS. The crest factor reads 16.2 dB before and after, which is how you can tell the shape didn’t change.
Audacity has this as its Normalize effect, which aims for −1.0 dB by default. Ableton Live’s export window has a Normalize switch that raises the bounced file until its highest peak reaches full scale, and Logic’s Bounce window has a Normalize menu whose Overload Protection Only setting will turn a file down if it would clip, but never up.
Loudness normalisation
Loudness normalisation works the same way with a different measurement. It reads the file’s integrated LUFS, the loudness of the whole track measured the way ITU-R BS.1770 sets out, and moves the gain until that number lands on the target. Audacity’s Loudness Normalization effect does this to a file, and broadcasters use it to deliver to specs like EBU R128, which asks for −23 LUFS.
The place you’ll meet it most is streaming. Spotify, YouTube and Apple Music use loudness normalisation to set each track’s playback level, so everything comes out at about the same loudness. They don’t change your file; the gain happens as the track plays. The streaming loudness lesson goes through each service.
Loudness normalisation has one catch that peak normalisation doesn’t. Turning a track up to a loudness target can push its peaks past 0 dBFS, which is why Spotify only turns a quiet track up as far as its true peak allows, and stops at −1 dBTP.
What is limiting?
A limiter is a compressor set so hard and so fast that nothing gets past its ceiling. It watches the level all the time, and whenever a peak is about to go over, it turns the sound down just enough, for just as long as it needs to, then lets the level back up. Most use lookahead, delaying the audio by a few milliseconds so they can start turning down before each peak arrives.
You turn the sound up into the limiter, and it holds the peaks at the ceiling while the quieter parts rise. That’s the whole difference from normalising. Push the lab’s full mix 8 dB into a limiter with a −1 dBTP ceiling and the true peak stays just under −1 dBTP, exactly where the normalised version sat, but the mix now reads −10.3 LUFS instead of −14.9. At the same peak it’s 4.6 LU louder, because the limiter took up to 8.9 dB off the loudest hit and let everything else come up.
That loudness has a price. The crest factor, the gap between the peaks and the average level, falls from 16.2 to 11.6 dB, and you can hear it on the front of the kick and snare, which lose a little of their edge. The limiting lesson covers the controls and how far to push.
Why doesn’t normalising make my track louder?
Because peak normalisation is controlled by one sample. Whatever the loudest moment in the file is, a snare hit or one loud word, it decides how far the whole file can come up. If that peak is already near 0 dBFS, there’s nowhere to go, and a mix that’s been through a limiter is usually right at its ceiling already.
It also means two files normalised to the same peak can sound nothing alike. The acoustic guitar and the distorted drums in the lab both have a sample peak of −0.9 dBFS. The guitar reads −21.8 LUFS and the drums −13.6, so after peak normalising them to the same level they’re still 8.2 LU apart. A plucked guitar has a sharp spike at the start of each note and then rings away, while the drums have been squashed until there’s little space between the hits.
Loudness normalising the guitar with gain alone shows the opposite problem. To bring it up to −14 LUFS you’d add 7.8 dB, and its true peak would go to +6.9 dBTP. Exported to a 24-bit WAV, everything over 0 dBFS would be cut off, so you’d end up with a clipped file. Plain gain can’t make this guitar that loud.
A limiter can. With 10 dB of push the guitar reads −14.1 LUFS and its true peak stays just under −1 dBTP, because the limiter turns the loudest pluck down by 10.2 dB. The crest factor falls from 24.8 to 16.9 dB. Listen for the attack of each note getting softer, which is the cost of that loudness on a spiky recording.
Does limiting still help once Spotify normalises?
Streaming services come after your master, and most of them normalise by default. So the real question for a finished master is what the limiter buys you once Spotify has had its say.
For a mix that’s already about as loud as the target, almost nothing. The limited full mix from earlier reads −10.3 LUFS, so Spotify turns it down 3.7 dB to −14. The untouched mix, at −14.1 LUFS, stays where it is. Both now play at about the same loudness, and the master’s peaks reach only −4.8 dBFS against the original’s −0.2, so the original has the bigger hits.
For a quiet, spiky recording, the limiter does change what the listener hears. Spotify can’t turn the original guitar up at all, because its peaks already reach −0.9 dBTP and Spotify won’t go past −1. So the original plays at −21.8 LUFS while the limited version plays at about −14, 7.8 LU louder. Next to other songs in a playlist, the unlimited one would sound far too quiet.
So limit enough to get a track near the loudness of the music around it, and stop there. Past that point, every extra decibel of limiting gets taken back by normalisation and leaves you with flatter peaks for nothing. The LUFS lesson explains how the integrated figure is measured.
Does normalising change the sound?
Only its level. The balance and the dynamics stay exactly as they were, which you can check in the lab: when you move the Gain knob, every reading moves by the same number of decibels and the crest factor doesn’t budge.
The level still affects how you hear it, though. Louder sounds seem fuller in the bass and treble, so a normalised file can sound better than the original for no other reason. So a fair comparison needs both versions at the same loudness, which is what the Spotify button gives you.
The numbers inside the file are recalculated when you normalise, and each one gets rounded. In a 24-bit or 32-bit float file that rounding sits more than 140 dB below full scale, far under anything you could hear. Normalise a 16-bit file several times over and the rounding builds up slowly, which is one reason to work at 24-bit or higher and leave the move to 16-bit, with dither, until the end.
When to use each
| Job | Normalise | Limit |
|---|---|---|
| A recording came in far too quiet | Yes, to a peak well under 0 dBFS | Not needed |
| Compare two mixes or a reference fairly | Yes, by loudness | No |
| Make a finished master as loud as other releases | Can’t go past its peak | Yes, last on the master |
| Deliver to a loudness spec such as −23 LUFS | Yes, to the target | Only if the peaks end up too high |
| Stop an export from clipping | Yes, turn it down to −1 dBTP | If you want to keep the loudness |
| Level a pile of one-shots for auditioning | Yes | No |
Most of the time the choice comes down to whether there’s room above the peaks. If there is, normalise or turn up the gain, because a volume change costs nothing and leaves the sound exactly as it was. If there isn’t, and the track still needs to be louder, that’s what a limiter is for.
How to make a quiet track louder, step by step
- Measure it firstPut a loudness meter on the master and play the whole track. Note the integrated LUFS and the true peak.
- Use the headroom you haveIf the true peak is well under −1 dBTP, raise the gain, or normalise to −1 dBTP, until it touches. That’s free loudness, and the sound doesn’t change.
- Limit what’s leftIf it’s still too quiet, add a limiter last on the master with its ceiling at −1 dBTP and true peak detection on. Raise its input until the loudest moments show 2 to 3 dB of gain reduction, then go further only if you need to.
- Compare at the same loudnessTurn the limited version down until it measures the same LUFS as the unlimited one and flip between them. If the limited version sounds worse at the same level, you’ve gone too far.
- Export without normalisingLeave any Normalize option in the export window switched off. Peak normalising to 0 dBFS after a limiter would lift your −1 dBTP ceiling straight back to full scale.
Common mistakes
Normalising everything to 0 dBFS before mixing
With every recording peaking at full scale there’s no headroom left. Add a few tracks together and the master goes over, and you’ll spend the mix pulling faders down. Keep raw tracks peaking well below 0 dBFS; the gain staging lesson has the numbers.
Normalising to 0.0 dBFS and calling it safe
A peak normaliser usually reads sample peaks, and the waveform between two samples can go higher than either of them. The lab’s bass loop shows it: its highest sample is −0.5 dBFS and its true peak is +0.03 dBTP. Those in-between peaks can distort once the file is turned into MP3 or AAC, so normalise to −1 dBTP, or about −1.5 dBFS if your editor only reads sample peaks.
Ticking Normalize on export after a limiter
The limiter held the peaks at −1 dBTP, and the export option turns the file up until its highest sample hits 0 dBFS. That undoes the margin you set and can add true peaks above full scale. Let the limiter set the final peak level and leave export normalising off.
Normalising stems or mix tracks one by one
Each file gets a different gain, so the balance you mixed is gone when someone loads them together. Send stems exactly as they came out of the session.
Limiting harder to beat streaming normalisation
A master at −8 LUFS plays at the same loudness on Spotify as one at −14; it just has smaller peaks. If a track sounds quiet next to others there, compare it with a reference at matched loudness and listen for what’s different in the mix.
Train your ears
The quiz under the lab plays two versions of a loop with the screens covered and normalisation off, and asks which has the higher integrated loudness. One is often limited and the other isn’t, so the louder one can be the one with the smaller peaks, which is exactly what a peak normaliser would miss. Eight rounds take a few minutes, and your best score is saved in your browser.
Questions people ask
Does normalising make audio louder?
Only if there was room above its highest peak. Peak normalising turns the whole file up until that peak hits the target, so a file peaking at −9 dBFS gets 8 dB louder when you normalise to −1 dBFS, and a file already peaking at −1 dBFS doesn’t change at all.
Is normalising the same as compression?
No. Normalising applies one fixed gain to the whole file, so the loud and quiet parts move together. A compressor, and a limiter, which is a very fast compressor with a very high ratio, change the gain from moment to moment and turn the loud parts down more than the quiet ones.
Should I normalise to 0 dB or −1 dB?
Use −1 dBTP if your editor offers true peak normalisation, or about −1.5 dBFS if it only reads sample peaks. A file normalised to 0 dBFS can have true peaks above full scale between the samples, and those can distort once the file is turned into MP3 or AAC.
Should I normalise before mastering?
There’s no need. Normalising only moves the level, and the mastering engineer, or your own limiter’s input gain, will set that anyway. Send the mix with its peaks under 0 dBFS and nothing on the master that you haven’t agreed on.
Does normalising reduce audio quality?
Not in any way you’ll hear. In a 24-bit or 32-bit float file, the rounding it adds sits more than 140 dB below full scale. The one real side effect is that any hiss or hum comes up by the same amount as the music.
Does Spotify limit my track?
Only in its Loud setting, which aims for −11 LUFS and uses a limiter to make room when it turns a quieter track up. In the default setting Spotify normalises with plain gain: it turns loud tracks down and turns quiet ones up as far as their peaks allow, without changing their shape.