Given a Replay Gain tagged file, at playback time there's usually the choice between applying the album gain or the track gain. Both have their pros and cons but in general neither strategy seems to be perfect.
If we consider album gain as the "natural strategy" because it preserves the loudness difference between an album's individual tracks we may consider track gain as some kind of "deviation" from the "natural strategy" aiming to correct some flaws of the "natural strategy".
Viewed this way the question rises why there shouldn't be a measure allowing us to control the "deviation" in a smooth way from 0% (album gain) to 100% (track gain)? The latest version with the FFSoX Player plugin for WA (http://in-ffsox.sourceforge.net/) implements the idea:
(http://in-ffsox.sourceforge.net/images/in-ffsox-0.5.0-part.png)
What do you think about it? Do you find it useful?
That's a solution to a problem that I don't know exists. Album gain = preserve the integrity of an album; makes sense when listening to albums. Track gain: make everything sound as loud; makes sense when shuffling (or picking single tracks from various albums). What other playback modes are there that would benefit from your solution?
That's a solution to a problem that I don't know exists.
That may be true for you, but for sure not for me.
Album gain = preserve the integrity of an album; makes sense when listening to albums.
Agreed. That's why I call it the "natural strategy".
Track gain: make everything sound as loud; makes sense when shuffling (or picking single tracks from various albums).
Not agreed. The goal is not to have everything as loud as possible but to have everything appearing as equal loud, and that's the crux.
What other playback modes are there that would benefit from your solution?
I had the idea last week and implemented it immediately. Having listened for some days to a shuffle with strategy "50%", I have the impression that it improves dramatically the strategy "track" == "100%".
Not agreed. The goal is not to have everything as loud as possible but to have everything appearing as equal loud, and that's the crux.
(emphasis mine) Yes, that's what I meant. I didn't say "as loud as possible", I said "as loud". I should have said "equally loud". But that's exactly what track gain does.
I think it's a good idea, thought I think there are too many choices. 25%, 50% and 75% should be adequate, 20% steps will definitely be adequate.
But that's exactly what track gain does.
It claims that it does, but it doesn't in practice.
I think it's a good idea, thought I think there are too many choices. 25%, 50% and 75% should be adequate, 20% steps will definitely be adequate.
I think a more appropriate GUI control would have been a slider. Having a drop down box was the easiest way to have an implementation without modifying the existing GUI (the drop down box was already there).
Maybe it doesn't work so well, but I have a hard time believing this idea is going to provide any help with that. Rather, I like this idea for reducing the amount of dynamic difference between tracks when shuffling. I've been vocal about my utter loathing of track gain in the past, but sometimes I think maybe the differences between quiet tracks and loud tracks in the context of shuffling may be a bit too large in some cases when using straight album gain.
It claims that it does, but it doesn't in practice.
Can you demonstrate, even empirically, how it doesn't? I mean, the algorithm is the same for both album and track gains. Why would track gain fail where album gain succeeds?
I think a more appropriate GUI control would have been a slider. Having a drop down box was the easiest way to have an implementation without modifying the existing GUI (the drop down box was already there).
I would still limit the number of steps on the slider. I don't believe such resolution is useful, let alone warranted. If people are going to use the function in a conversion process a non-detented slider can be problematic. They may adjust the slider between conversions and not remember where they had it previously set.
Can you demonstrate, even empirically, how it doesn't?
The forum has had many a testimonial on RG doing less than a perfect job. I'll again offer up how my Beatles tracks seem to be a good 4-6 dB louder to my ears than an average non-Beatles track in my collection (vocals are far too punchy).
I don't think this will provide a solution to that problem however.
That said, maybe you guys are discussing some other, more nuanced problem than RG track gain really doesn't work as advertised (same can be said for album gain and to an even greater extent, but I fear that would be going too far off-topic).
Can you demonstrate, even empirically, how it doesn't?
As you probably know the perceived loudness is a highly subjective matter. Existing algorithms (e.g. ReplayGain or ITU BS.1770) have demonstrated to be very good approximations. However, listening to a shuffle I (subjectively) have the impression that from time to time there's a "loudness jump" exactly in the context as described by greynol:
but sometimes I think maybe the differences between quiet tracks and loud tracks in the context of shuffling may be a bit too large in some cases when using straight album gain.
I have to admit, I never shuffle or make custom playlists that aren't made of entire albums. Maybe Replaygain and co. work better over longer pieces music (albums) where anomalies are comparatively small (e.g. the length of an average track). If so, it would be interesting to revise those algorithms.
As for your solution being useful or not: it's obviously useful to you. That should be enough for you to keep and implement that feature wherever you can (unless you don't use your own products!).
it's obviously useful to you. That should be enough for you to keep and implement that feature wherever you can (unless you don't use your own products!).
One thing's for certain: that's the driving force for everything I'm doing You can be sure, in the first place I'm writing these programs for myself and I'm using them extensively myself. Hopefully it's useful for somebody else too ...