Re: increasing / normalizing loudness for large numbers of wav (David Robinson )

Subject: Re: increasing / normalizing loudness for large numbers of wav
From:    David Robinson  <david(at)ROBINSON.ORG>
Date:    Wed, 28 Jul 2004 04:22:10 -0700

On Tue, 27 Jul 2004 01:00:39 -0700, Timothy Justus wrote

> Could anyone recommend a good, free program that I
> could use to increase and
> equate the loudness of a large set of wav files (each
> containing a spoken
> word)? Those that I've been experimenting with either
> require working with
> one file at a time (e.g. audacity) or introduce large
> distortion into my files.

Matching the perceived loudness of sounds isn't a
trivial task.

Matching or maximising the peak amplitude value in
several audio files will not match their loudness.
Matching the RMS level will get you closer, but can
still be way off.

I developed a rough-and-ready algorithm called
ReplayGain which is designed to match the loudness of
music tracks by taking the 95th percentile of a
sequence of frequency weighted 50ms RMS measurements.
It usually works quite well, but may not work too well
for single words! You can try it on wave files using
this utility:

http://members.home.nl/w.speek/wavegain.htm

You could use a more advance algorithm for matching
loudness - but maybe in your application you don't want
an exact loudness match? Maybe the perceived loudness
of some words should be slightly different from others,
and you want to maintain that?

Hope this helps.

Cheers,
David.