Currently, normalization seems to adjust the volume of every track by itself, regardless of where it’s playing from. This works for the most part, but it has the consequence of messing with the dynamics of full albums. It’s especially jarring in albums with gapless transitions between tracks, where Deezer’s normalization can cause sudden volume changes where none was intended, but it affects all albums to some extent.
To remedy this, Deezer should calculate a separate “album loudness” to be used when a user begins playback from an album. That way, the track-to-track relative volume differences/dynamic range that the artist intended to be heard can be preserved, while the current behavior can remain for other playback sources (playlists, favorites, etc.) This is also in keeping with established standards. The EBU R 128 S2 (Loudness in Streaming) recommends that:
“especially when individual tracks are streamed as separate elements (for example, music
services) additional metadata may be used to ensure faithful reproduction of the artistically
intended relationship between programmes. These can indicate, for example, the loudest
track of a music album, the loudest movement of a classical symphony or the loudness of
speech (“album normalisation”, “anchor-based normalisation”).“
