Normalization analyzes an audio file and automatically adjusts the overall gain so that it matches a chosen target value. Because it basically applies the same amount of gain to everything, normalization on its own does not change the relative difference between loud and quiet parts.
Peak normalization
Peak normalization finds the highest sample peak and raises or lowers the whole file so that it reaches a target such as −1 dBFS or 0 dBFS. The maximum values line up, but two tracks with different density will not necessarily end up at the same perceived loudness.
Loudness normalization
Loudness normalization measures the perceived loudness of the whole file with a metric such as integrated LUFS and matches it to a specified value. It is used for playback level adjustment on streaming services and to reduce level differences across several videos or audio files. It uses how the material sounds overall, rather than its peaks, as the reference.
How it differs from a compressor
Normalization applies, in principle, the same gain across the whole file. A compressor or a limiter applies different amounts of gain over time, for example only where the signal exceeds the threshold. So the former preserves the relative dynamics, while the latter can change the dynamic range.
Watch true peak as well
If you set the sample peak to 0 dBFS, no room is left before true peak exceeds the ceiling during playback or MP3 and AAC conversion. Pushing up to a loudness target can also make the peaks hit the ceiling first. Using a limiter in that case means the process is no longer simple normalization.
Choose the reference that fits the purpose
A peak reference suits trimming the maximum peak of a single file, and a LUFS reference suits matching the perceived loudness of several files. If the destination has a spec, give priority to its loudness value and its true peak ceiling. The tool for matching the loudness of multiple files lets you adjust them against a common integrated LUFS target.
