Peak normalisation against loudness normalisation
Peak normalisation raises a file's gain until its loudest sample reaches a chosen ceiling. It guarantees no clipping and is trivial to compute — but it says nothing about perceived loudness, because a track with one brief transient peak and a generally quiet body will be normalised on that peak and still sound quiet.
Loudness normalisation measures perceived loudness instead, using the ITU-R BS.1770 algorithm, and adjusts gain so that measurement hits a target. The unit is LUFS, Loudness Units relative to Full Scale, and it applies frequency weighting that approximates human hearing's greater sensitivity in the midrange. This is what actually makes a set of tracks sound consistent.
The related measurement is true peak, measured in dBTP. Digital samples are points on a waveform, and reconstruction between them can exceed the highest sample value — so a file peaking at exactly 0 dBFS can clip in a digital-to-analogue converter or after lossy encoding. Leaving a ceiling around −1 dBTP avoids this.
Platform targets and why loudness wars ended
Streaming platforms normalise playback to their own targets, which changed the economics of mastering. Spotify targets around −14 LUFS integrated, YouTube around −14, and Apple Music around −16. A master delivered louder than the target is simply turned down on playback.
That is why the loudness war — decades of progressively more compressed, louder masters competing for perceived impact — lost its purpose. A track crushed to −6 LUFS is attenuated by 8 dB to meet the target, arriving at the same loudness as a dynamic master but with its transients permanently destroyed. The dynamic version now sounds better and no quieter.
Broadcast has its own standards and they are enforced: EBU R128 specifies −23 LUFS in Europe and the US CALM Act mandates −24 LKFS, both introduced specifically to stop advertisements being louder than programming. Podcasts conventionally sit around −16 LUFS mono or −19 stereo.
Normalising a set of files
The common mistake when processing an album or a podcast series is normalising each file independently to the same target. That flattens the intentional relationships between tracks — a deliberately quiet interlude is raised to match the loudest song, destroying the dynamic arc.
Album normalisation applies a single gain adjustment across the whole set, calculated from the collection as a whole, so relative levels are preserved while the set as a whole hits the target. Most streaming platforms offer this as an option, and it is the right choice for anything intended to be heard in sequence.
Normalisation is gain adjustment only. It does not fix a recording with a poor noise floor, inconsistent microphone technique or wildly varying levels within a single file — raising the gain raises the noise with it. Those require compression or manual level rides, which are different operations. For removing dead air rather than adjusting levels, the silence remover handles that separately.