Loudness normalization is the process of adjusting the gain of audio programs so that they play back at the same perceived loudness, measured against a target in LUFS. In mastering and distribution, it replaces the older practice of matching peak levels, which allowed heavily compressed material to sound much louder than dynamic material.
Broadcasters, streaming services and playback software apply loudness normalization automatically. As a result, the integrated loudness of a master, rather than its peak level, decides how loud it will actually sound to most listeners, which has changed how mastering engineers balance level against dynamics.
How loudness normalization works
Loudness normalization works by measuring the integrated loudness of a program and applying a single gain offset that moves it to the target value. If a song measures -8 LUFS and the target is -14 LUFS, the system turns it down by 6 dB; the shape of the waveform and its internal dynamics stay unchanged.
The measurement follows ITU-R BS.1770, which applies a K-weighting filter that approximates human hearing sensitivity, averages the energy over time and uses gating to ignore silence and very quiet passages. The result is expressed in LUFS (loudness units relative to full scale) or the equivalent LKFS, where one loudness unit corresponds to one decibel.
The table compares loudness normalization with the older approach of peak normalization.
| Property | Peak normalization | Loudness normalization |
|---|---|---|
| Reference value | Highest sample or true peak | Integrated loudness in LUFS |
| Matches perception | Poorly | Closely |
| Effect on compressed masters | They sound louder | They are turned down |
| Typical use | Editing and file preparation | Broadcast, streaming and playlists |
| Risk | Large loudness jumps | Clipping if gain is raised |
Common loudness targets
Loudness targets differ by medium and standard, which is why the same master can be turned down by different amounts on different services. Broadcast standards are strict regulations, while streaming targets are platform choices that may change and are not always fully documented.
| Context | Typical target | Basis |
|---|---|---|
| European broadcast | -23 LUFS | EBU R128 |
| United States television | -24 LKFS | ATSC A/85 |
| Major music streaming, default | Around -14 LUFS | Platform policy |
| Some music streaming players | Around -16 LUFS | Platform policy |
| Podcasts and spoken word | Commonly -16 to -19 LUFS | Industry practice |
Broadcast standards also set a maximum true peak, for example -1 dBTP in EBU R128, and specify a permitted tolerance. Streaming targets are usually applied as playback gain rather than by altering the delivered file, and users can sometimes switch normalization off or choose louder or quieter presets.
Track mode and album mode
Normalization can be applied per track or per album, and the choice affects how intended level differences survive. In track mode, each song is moved to the target independently; in album mode, the whole album receives one gain offset so a quiet ballad stays quieter than a loud opener.
| Mode | How gain is set | Best suited for | Side effect |
|---|---|---|---|
| Track mode | Each track reaches the target | Shuffled playlists and radio | Ballads become as loud as anthems |
| Album mode | One offset for the whole album | Albums played in order | Quiet tracks stay quiet |
| Broadcast program | Whole program measured | Television and radio shows | Internal mix balance preserved |
Many players use album mode when a listener plays an album from start to finish and track mode for mixed playlists. Mastering engineers therefore set relative loudness between songs deliberately, knowing it survives only when album mode is active.
What loudness normalization means for mastering
Loudness normalization means that extra loudness beyond the target brings no playback advantage on normalized platforms. A master at -7 LUFS and a master at -13 LUFS will play at nearly the same level, but the quieter one keeps more transient detail and punch.
The table below summarizes practical decisions for mastering engineers working with normalized delivery.
| Decision | Recommended approach | Reason |
|---|---|---|
| Integrated loudness | Choose by genre, not by maximum level | Excess loudness is removed anyway |
| True peak ceiling | About -1 dBTP, lower for very loud masters | Lossy encoding can create overs |
| Dynamic range | Keep transients where the music needs them | Normalized playback rewards dynamics |
| Quiet masters | Check whether the platform raises gain | Raising gain may trigger a limiter |
| Reference comparison | Compare at matched loudness | Avoids louder-is-better bias |
Some services raise quiet tracks toward the target only when enough peak headroom exists, while others apply a protective limiter. A very dynamic master with high peaks may therefore remain slightly below the platform target, which is usually acceptable.
Limitations and misconceptions
Loudness normalization equalizes average loudness, not tonal balance or density, so masters can still feel different at the same LUFS value. A bright, dense master may seem more forward than a dark, open one even when both measure exactly the same.
- Normalization is not compression: it applies one static gain change and leaves dynamics untouched.
- Integrated loudness is not short-term loudness: a track with quiet verses and loud choruses may measure lower than its choruses suggest.
- Targets are not mastering requirements: delivering exactly -14 LUFS is not mandatory for music streaming.
- Users may disable normalization, so masters should also sound acceptable without it.
Measurement also has known weaknesses. K-weighting underestimates the perceived loudness of some very bass-heavy material, and the gating thresholds defined in EBU R128 and BS.1770 make long silent sections irrelevant to the reading, which is intended but surprises some users.
Stripped to the essentials
Loudness normalization is an automatic volume leveler for music, shows and podcasts. Before it existed, listeners often had to reach for the volume control when one song was much louder than the next. Now the player measures how loud each piece sounds to a person and quietly turns the louder ones down.
The practical effect is simple: making a recording extremely loud no longer makes it stand out. It only gets turned down, and it may lose some of its energy in the process.
Where it is applied
A mastering engineer receives a rock album and prepares two versions for comparison. The first version is pushed to -7 LUFS integrated with heavy limiting; the second stays at -11 LUFS with about 2 dB of limiting and a true peak of -1 dBTP.
When both are played through a streaming player normalizing to around -14 LUFS, the first version is lowered by 7 dB and the second by 3 dB. Both now play at the same level, but the second version keeps noticeably sharper drums and wider dynamic contrast between verses and choruses.
The band approves the second version for release, and the album also plays back consistently in album mode, where the quieter closing song keeps its intended lower level.