I haven't solved the problem of balancing
MP3s, but I have thought about it.

I want to apply a single scalar amplification
to balance every track so that I no longer
need to adjust the volume knob on my digital
player.  I do not want a running automatic
gain control (AGC) that compresses the
dynamic range.  I want the typical amplitude
to be relatively constant, but without any
danger of really loud bursts.  You'd think
that scaling by peak amplitudes might be
better.  I've tried that, and it doesn't
work.

I use "lame" to convert WAV's to MP3s.  It
includes the value of the peak amplitude as a
"Replay Gain" in the MP3 header.

http://en.wikipedia.org/wiki/Replay_Gain

Nevertheless, tracks still have variable
volume on my players.  This might be because
every player I've tried, including software,
ignores "Replay Gain," but I think the real
problem is clipping.

As years go by, CD's have been increasingly
clipped, with waveforms squared off, to make
the typical volume higher.  Clipping is a
stupid filter that destroys signal and
dynamic range, without changing the power of
your amplifier.  But you can't escape
clipping anymore.

http://www.mindspring.com/~mrichter/dynamics/dynamics.htm

How do you undo this kind of damage?  Clearly
you can't scale by the peak amplitude
anymore.  I'd prefer an root-mean-square
(RMS) amplitude, but again, that allows for
ear-destroying bursts in tracks with a real
dynamic range.  I don't want to add clipping
to good tracks, or to apply an AGC.  I just
want to scale down clipped tracks so they
aren't playing at the max all the time.  I
think I want to multiply the RMS amplitude by
some factor, and use that value as "Replay
Gain" if larger than the maximum of the
entire track.  So what's the right factor?

"Lame" can also apply a scalar gain to
encoded amplitudes, but not automatically.  I
would only dare scale down amplitudes since
scaling up might cause more clipping.

Bill Harlan, April 2007