Skip to main content
Topic: ReplayGain AlbumGain - is mean average the wrong way? (Read 267 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

ReplayGain AlbumGain - is mean average the wrong way?

I did search for previous threads, but discussion didn't seem to be going in this direction.  If there is another thread discussing my point, I sincerely apologise.

We have ReplayGain (R128 algorithm) which is just fine for tracks.  It's an integrated LUFS scan, with an adjustment to make any track meet a I-loudness of -18 LUFS.  That's fine by me.

What I am questioning is the current implementation of Album Gain.
Currently, all tracks in an album (either by tag or by "scan as single album") are scanned for I-LUFS and the mean average is taken (add all RG_track values together and divide by number of tracks).
I think a better solution would be to take the track with the greatest level of reduction and use that for the Album Gain value.

My reasoning is that I am listening to a collection of tracks picked from albums and the levels seems a bit skewed.  I do not like RG track gain mode as it doesn't attempt to preserve the intent of each track.  A thick mix rock track shouldn't be adjusted to the same I-LUFS as a soft acoustic guitar and female voice in my opinion.

I took it upon myself to copy out the meta tags and set the entire album gain to that of the 'loudest' track (the one with the greatest -ve RG track value) and things work a lot better.  The listening experience is much smoother.

What are people's thoughts on this?
If anything, a switch would be nice to allow a choice of either mean average of loudest track as the album gain value.

Thanks.

Re: ReplayGain AlbumGain - is mean average the wrong way?

Reply #1
The current Album Gain method is supposed to be identical to scanning the entire album as a single track. Any math to blend multiple scans together is designed solely for the purpose of multi-threaded scanning optimization, so the scanner may scan one track per thread, instead of having to scan the entire sequence of tracks in only one thread.

Re: ReplayGain AlbumGain - is mean average the wrong way?

Reply #2
That is interesting information.

Thinking about this a little more, I wondered what good may come of looking more towards maximum Momentary or Short-term loudness, and using that as a point of reference.  The remark I think of in response to my own point is that if an album is quiet-ish with a particularly loud short term peak in loudness here and there, then is it better to have it roughly the same level as a fairly non-dynamic album or to turn it all down just because of a few short periods of relative loud?
Another way to say what I just tried, is to take an example of a orchestral music with a ridiculously loud section - does that short section dictate the loudness of the entire thing, or is that allowed to jump out of the comfort zone for a short while for wonderful dynamic impact?
I suppose my answer would be yet another switch/option.

I am aware of the old RepalyGain algorithm, and R128 implementations have just changed the definition of how loud, on average, a track or album is.  So, I can see that where I am thinking is quite a different a radical change to things.

I will have to edit my RG tags myself and see how things fair for me, doing things as I original proposed.  So far, it seems to be working nicely, but I am bound to find issues here and there.

Re: ReplayGain AlbumGain - is mean average the wrong way?

Reply #3
My reasoning is that I am listening to a collection of tracks picked from albums and the levels seems a bit skewed.  I do not like RG track gain mode as it doesn't attempt to preserve the intent of each track.  A thick mix rock track shouldn't be adjusted to the same I-LUFS as a soft acoustic guitar and female voice in my opinion.

Whether it be a rock track or just a vocal and acoustic guitar, isn't adjusting each track to the same apparent volume (Track Gain), the whole point of using ReplayGain?

Re: ReplayGain AlbumGain - is mean average the wrong way?

Reply #4
Track gain is no good. As was pointed out by Frank Klemm, it will sometimes boost quiet titles and their noise too much. For example spoken intros or short instrumental segments that are separated on their own track. These titles could be appended or prepended to a music track, and would take its gain then. Album gain does well to average out these differently indexed albums.

An acoustic recording isn't guaranteed to have loud tracks to force down this hybrid gain. I think Momentary loudness is too short to base the gain upon. Not sure about Short Term. You may find that Compilations get attenuated too much if they include one or more tracks that are loud for no good reason, because they were brought in from a different source and not levelled. This hybrid gain (loudest track) was chosen by one streaming website, as described one of Ian Shepherd's interviews (which I can't find now). But I don't believe it is right.

I sorted my collection by the difference between track and album gain, and I see quite a few pop/rock interludes that are boosted by around +14. (Simon and Garfunkel - Voices of Old People, The Corrs - Erin Shore, Sandra - The Second Day). At the other extreme, where a track is louder than average, I see a difference of only around -4 dB, a few Compilations have -6.

 
SimplePortal 1.0.0 RC1 © 2008-2019