Normalization

Audio normalization is the application of one constant gain change to a whole file or clip so that it reaches a target level, called the norm. The target can be a peak level (for example −1 dBFS) or a loudness level (for example −14 LUFS).

Because the same gain is applied everywhere, normalization makes audio louder or quieter without changing its dynamics. That is the key difference from compression and limiting.

What is normalization?

Think of normalization as an automatic volume knob. The software measures the audio, works out how far it is from the target, and turns the whole thing up or down by exactly that amount. A file whose loudest peak is at −6 dBFS, normalized to 0 dBFS, is simply raised by 6 dB. The distance between the loudest and quietest moments stays exactly the same.

The Wikipedia entry on audio normalization describes it as applying a constant amount of gain to bring the amplitude to a target level, leaving the signal-to-noise ratio and relative dynamics unchanged. In broadcast, the ATSC A/85 recommended practice uses loudness normalization to establish and maintain a target program loudness of −24 LKFS, so viewers are not hit with sudden volume jumps.

Two kinds of normalization

• Peak normalization adjusts gain so that the highest peak lands on the target. It guarantees the file will not exceed a certain level, but says nothing about how loud it sounds.
• Loudness normalization adjusts gain so that the measured loudness (in LUFS or LKFS) lands on the target. It makes different files sound equally loud, which is what broadcasters and streaming services use.

Not the same word in music theory

In music theory, particularly pitch-class set theory, “normal form” or “normal order” describes the most compact arrangement of a set of pitch classes. It has nothing to do with audio level; the two uses only share the general idea of putting data into a standard form.

The history of normalization

Peak normalization in the first digital editors

Digital audio editors of the late 1980s and 1990s made peak normalization a standard function: scan the file for its highest sample and scale everything so that sample sits at 0 dBFS. It was simple and exact, and it exposed a problem. Two files normalized to the same peak could sound very different in loudness, depending on how much of their energy sat close to that peak.

The loudness war and the need for a better measure

Because CDs were effectively peak-limited to 0 dBFS, the only way to sound louder was to raise the average level with heavy compression and limiting, feeding the loudness war. Normalizing by peaks did nothing to solve it; normalizing by perceived loudness would.

ReplayGain and the MP3 era

As digital music libraries grew in the MP3 era, jumping levels between tracks became a daily annoyance. In 2001, David Robinson proposed ReplayGain, which analyses the loudness of each track (and album) and stores the result as metadata so that players can adjust playback volume. It was one of the first widely used loudness normalization systems for music.

Broadcast standards

In 2006 the ITU published ITU-R BS.1770, the loudness measurement behind LUFS, building on research by Gilbert Soulodre. The EBU’s PLOUD group then produced EBU R128 (2010) with a −23 LUFS target, and in the US the CALM Act (2010) turned the ATSC A/85 practice into the basis for regulating commercial loudness on television.

Streaming

Streaming services adopted loudness normalization for music playback, and in 2021 the Audio Engineering Society published AES TD1008, recommendations for loudness in internet streaming, including album normalization and separate targets for music and speech.

How normalization works

Sample peak normalization

The software finds the highest absolute sample value, calculates the difference to the target in decibels, and multiplies every sample by the matching factor. The calculation is trivial: target level minus current peak equals the gain to apply. If the peak is at −6 dBFS and the target is −1 dBFS, the gain is +5 dB.

True peak normalization

Sample values are not the whole story. When audio is converted back to analog, or encoded to a lossy format, the reconstructed waveform can rise between the samples. True-peak measurement, specified in ITU-R BS.1770, oversamples the signal (at least 4× for 48 kHz audio) to estimate these inter-sample peaks. Normalizing to a true-peak target such as −1 dBTP is safer than normalizing to a sample-peak target, especially for files that will be encoded.

RMS normalization

Normalizing by RMS uses the average power of the signal instead of its peaks. It tracks perceived loudness better than peak normalization, but it treats all frequencies equally. A track with heavy sub-bass measures as “loud” on RMS even though the ear hears relatively little of that energy.

Loudness (LUFS) normalization

Loudness normalization measures integrated loudness with the BS.1770 method: K-weighting to model hearing, 400 ms measurement blocks, and gates at −70 LUFS (absolute) and 10 LU below the average (relative) so silence and quiet passages do not distort the result. The difference between that measurement and the target becomes one fixed gain offset for the whole file. The details of the measurement are explained on the loudness page.

What normalization does not change

Because it is a single multiplication, normalization does not change the crest factor, the dynamic range, the balance between instruments or the signal-to-noise ratio. Noise goes up or down with the music by exactly the same amount. Apart from rounding when the result is saved at a fixed bit depth, normalization can only damage audio if the gain pushes peaks over 0 dBFS, or if a system raises a quiet track and then applies a limiter to stop it clipping.

How streaming services normalize

Track normalization vs album normalization

This distinction matters a lot for artists and mastering engineers:

• Track normalization measures each song separately and adjusts each to the target. It is used for playlists and shuffle.
• Album normalization applies one gain offset to a whole album, usually calculated so the album sits at the target overall or its loudest track does. A quiet interlude that was mastered 6 dB below the single stays 6 dB below it. Spotify normalizes an album as a whole when it is played as an album, and AES TD1008 recommends album normalization for music.

Platform behavior

Platform	Normalization	Notes
Spotify	−14 LUFS by default; −11 (Loud) and −19 (Quiet) for Premium	Album-aware. Positive gain on quiet tracks keeps 1 dB of headroom; a limiter is applied only in the Loud setting (Spotify for Artists).
Apple Music	−16 LUFS (Sound Check)	Sound Check measures in LUFS and is enabled by default on new installations of iOS and macOS (MeterPlugs).
YouTube, Tidal, Amazon Music	Around −14 LUFS (widely measured and reported)	Largely turn loud material down; behavior for quiet material varies.
AES TD1008 (recommendation)	−16 LUFS for music, −18 LUFS for speech	Recommends album normalization for music; in album mode the loudest track may play somewhat louder.

Platforms change their implementations from time to time, so check the current documentation before an important release. The principle stays the same: a platform’s target is where it plays your music, not a level you must master to.

What this means for mastering

A pop or EDM master at −7 LUFS and a jazz master at −16 LUFS can both be perfect. On Spotify’s default setting, the first is turned down by about 7 dB and keeps its dense, saturated character; the second may be turned up only as far as its peaks allow. Master for the music, then check how it sounds after normalization, using a loudness meter or a normalization preview tool.

Normalization in practice

Where normalization is useful

• Sound effect and sample libraries: batch-normalizing to a consistent peak (for example −3 or −6 dBFS) so assets arrive at predictable levels.
• Speech and podcasts: loudness-normalizing episodes and segments to a delivery target such as −16 to −19 LUFS for streaming, after editing and cleanup.
• Broadcast and film delivery: loudness-normalizing to −23 LUFS (EBU R128) or −24 LKFS (ATSC A/85), usually together with a true-peak limiter.
• Quick level checks: bringing a very quiet recording up so you can hear and edit it, before proper gain staging.

Where it causes problems

• Normalizing clips in a mix: peak-normalizing every track to 0 dBFS gives you tracks that are far too hot for plugins designed around lower operating levels, and leaves no headroom. Use clip gain and faders for gain staging instead.
• Normalizing before noise reduction: raising a noisy recording raises the noise with it. Clean first, then set the level.
• Normalizing each dialogue clip separately: a whisper and a shout end up at the same peak level, which flattens the natural conversation. Level by ear or with loudness tools on the whole program.
• Normalizing your mix to 0 dBFS before sending it to mastering: mastering engineers usually want headroom. Deliver the mix at its natural level unless asked otherwise.

Normalization in your DAW

DAW	How to normalize	Destructive?
Pro Tools	AudioSuite > Other > Normalize; loudness tools via AudioSuite plugins	Renders a new file
Logic Pro	Audio File Editor (formerly Sample Editor) > Functions > Normalize; region gain for manual changes	Audio File Editor edits the file
Ableton Live	Normalize option in the Export Audio/Video dialog; Clip Gain for clips	Export only
FL Studio	Edison > Normalize; Normalize checkbox in the sampler channel settings	Edison edits; channel option is non-destructive
Cubase / Nuendo	Audio > Direct Offline Processing > Normalize, using maximum peak or loudness as the reference	Non-destructive and editable later
Fender Studio Pro (formerly Studio One)	Audio > Normalize Audio on selected events	Non-destructive event gain
Bitwig Studio	Normalize in the audio event menu inserts a gain expression point towards 0 dB (Bitwig user guide)	Non-destructive
Reaper	Item normalization by peak, RMS or LUFS; Batch File Converter for many files	Non-destructive item volume

Tools for normalization

• iZotope RX Loudness Control: offline loudness normalization with presets for broadcast and streaming deliverables and a built-in true-peak limiter.
• Waves WLM Plus: a loudness meter that can calculate and apply the trim needed to reach a target.
• Auphonic: an online service widely used by podcasters for loudness normalization and levelling to standard targets.
• Broadcast processors such as Orban’s Optimod range: real-time loudness control for on-air signals.

Common mistakes and myths

• Myth: normalization ruins audio quality or dynamics. A correct normalization is a single gain change: dynamics, balance and tone are untouched, as long as the result does not clip.
• Myth: you must master to exactly −14 LUFS. That is a playback target on some platforms, not a delivery rule.
• Myth: normalizing to 0 dBFS is “safe”. A file normalized to 0 dBFS sample peak can exceed 0 dBTP after conversion or encoding. Use −1 dBTP or lower for delivery.
• Mistake: using normalization instead of gain staging. It sets a file’s level; it does not replace a sensible level plan through your mix.
• Mistake: forcing every album track to the same loudness. Album normalization exists precisely to preserve the differences between songs.

Normalization vs related terms

Comparison	The difference
Normalization vs compression	Normalization applies one static gain change; compression changes gain over time depending on the signal, reducing dynamic range.
Normalization vs limiting	A limiter stops peaks from crossing a ceiling; normalization just moves everything up or down.
Normalization vs clip gain	Clip gain is a manual level change chosen by the engineer; normalization calculates the gain automatically to hit a target.
Peak vs loudness normalization	Peak normalization targets the highest sample or true peak; loudness normalization targets perceived loudness in LUFS.
Track vs album normalization	Track normalization adjusts each song separately; album normalization applies one offset to the whole album.
ReplayGain vs Sound Check	ReplayGain is an open metadata standard from 2001; Sound Check is Apple’s playback normalization system.

Standards and research

Standard	Target	Domain
ITU-R BS.1770 (current revision -5)	Measurement method (LUFS/LKFS, true peak)	Global
EBU R128	−23 LUFS, maximum −1 dBTP	European broadcast
ATSC A/85	−24 LKFS	US television
AES TD1008 (2021)	−16 LUFS music, −18 LUFS speech, album normalization for music	Internet streaming and on-demand audio

Related terms

• Loudness: what loudness normalization measures, and how.
• Dynamic range: what normalization leaves untouched.
• Headroom: what you give up if you normalize everything to the top.
• Gain: normalization is simply an automatically calculated gain change.
• RMS: an older basis for level matching.
• Limiter: often used alongside loudness normalization to meet a true-peak limit.

FAQ