[Rhythmbox-devel] Replay gain

["Thomas Zander" <thomas e zander googlemail com>]

Good evening,

every now and then this mailing list faces a thread about RB's
replaygain support. There are a few bug reports out there, for example
http://bugzilla.gnome.org/show_bug.cgi?id=133881
http://bugzilla.gnome.org/show_bug.cgi?id=335073
http://bugzilla.gnome.org/show_bug.cgi?id=512788

In general, the consensus is that replaygain support can be called
decent if RB is able to just set the right gain without the user havin
to worry about it.
Especially tricky from RB's point of view is the question whether to
apply track gain or album gain for a particular song.
When we closely look at the different music playing scenarios, we
notice that album gain is actually rarely needed:
1) Mixed collection: Different artists, different songs from different
albums. Obviously we would choose track gain.
2) Different songs by the same artist but different albums: Surely
again track gain, as it does not make sense to apply a gain that has
been calculated from an average of songs that are not going to be
played
3) Songs from the same album in consecutive order: Not obviously track
gain, but also not necessarily album gain. Here is why:
3.1) Suppose the album is a collection of different songs, each one is
an "entity" of its own, the default case in pop music. Track one is
the single that has been sold 5 million times, the rest is more or
less boring, each of these songs has a second of silence at the
beginning and at the end. Track gain would be totally alright in this
case because there is no justification that one song on an album like
this should appear louder than a different one.
3.2) "Concept albums": Here we go. Consecutive tracks on the album
tell a story line that spawns beyond the borders of one song. Often
(and this is the kicker) tracks are supposed to crossfade into each
other. This is the one and only case where album gain makes sense,
because there are sequences that are shared by several tracks and they
are not supposed to spontaneously change the volume at some arbitrary
point that represents the track boundary on the CD.

Problem: How can RB know whether it just plays a file according to 3.1) or 3.2).
Probably it doesn't. Therefore, the logical and viable conclusion imho
should be that track gain is applied unless consecutive tracks from
the same album and artist are played. So, an automatic decision logic
would apply album gain in 3) and track gain in all other cases.

Comments? Suggestions?

Regards,
Riggs