Replay Gain - A Proposed Standard

Fixed Equal Loudness Curves aren't good enough!


7/11/2001, 9:01 pm

Tony wrote the following on the Monkey's Audio Message Board:

Of course you have to take some more stuff in to account.

a) you'd have to EQ the song as well,
since music is mixed and mastered to be heard at certain volumes(B.Spears-> kitchen radio conversation level to stomach thump of Guns'n'Roses). The Fletcher/Munson curve changes considerably at different phon ('loudness') levels.

Thus you'd have to take in to account what level the user has his reproduction system set, at which full scale digital signals hits him/her.

For this reason theatre reproduction systems(and the mixing stages) are calibrated, so it'll sound the same all over the place.

This is the only thing I don't see addressed in your proposal. The scan routine also needs to analyse the frequency content for possible meant playback levels of sound pressure levels(defined in dB SPL), or simply ask the user what this track is. Rock -> bass&treble a little lower, since it's meant to be played at higher volumes usually, and thus for people playing their music at a whisper, needs to have it's basic bass and treble frequencies boosted during playback.

The routine really needs to analyse this and take that in to account for the recommendation values it sticks in the file.


There's a sophisticated loudness measurement suggested by Brian Moore which takes the different Fletcher Munson curves at different levels into account all this. There's only one problem - it's about 1000x slower to compute. The accuracy gains are less than you'd expect.

Re: Some tracks needing to be louder than others. I don't think it's possible to build the intelligence in to detect this automatically (unless the relative levels are correct on a CD - then processing the whole CD at once as I've suggested will do the trick). So there will need to be human correction sometimes. However, I believe what I'm suggesting is a good start.

In response to such criticisms of the original proposal, I've come to the conclusion that storing two gains is the best solution. See Radio and Audiophile.