[MPlayer-dev-eng] [PATCH] audio filter volnorm2 - two-pass audio normalizer

[Vladislav Naumov]

Thanks to everybody who answered.
I'm convinced.
RMS normalization is better, since has more things in common with 
audible volume.
I'll try to implement it as (default) option.

What I want is to allow users to create not-too-bad MP3 rips from AC3
tracks. I understand that downmixing and re-encoding destroys
audio quality, so when quality IS important (nice effects, music videos
etc) I always just copy AC3 sound - and recommend it to complete
mencoding-newbies.

But in most cases audio quality is't an issue - 128kbps MP3 is just OK
for most drama/comedy movies.

So, I assume that liba52 downmixes to 5.1 to stereo as good as it is
possible. Now I need a normalizer - not perfect, maybe, but not too bad.
Something that works (with default settings) in most cases without
guessing cryptic values and "it depends on what sound do you have,
encode it 5 times with different settings, analyze results and pick
best one".
It's like deinterlacing filters - none of them is perfect, but -vf pp=lb
is not too bad.

Earlier I used to encode some sound, load it into CoolEdit/SoundForge,
find peak value (now I know what's wrong with it), and then re-encode
with vol="gain value". I didn't like it much, because it required
windows/wine to run. In Audacity 1.2.3 "Analyze" is still an empty menu.

That's why users keep asking for peak normalization - it is a simple way
to create a "mostly ok" sound.

And if users keep asking for peak normalization - and it is so simple
- maybe it will be better if peak normalization will be there?
With some kind of notice like "many people asked for that, so here it
is, but you better use RMS normalization (option bla-bla-bla)",
of course?