Subject: Re: [linux-audio-dev] decoding mp3 files
From: Paul Kellett (paul.kellett_AT_maxim.abel.co.uk)
Date: Sat Jun 08 2002 - 12:54:21 EEST
"Kristian Peters" <kristian.peters_AT_korseby.net> wrote:
>
> tim <tim_AT_ipd50abcd9.speed.planet.nl> wrote:
> > i would be interested to know which psyco-acoustic model
> > you are going to use to determine subjective quality.
>
> There're several ideas on that: I could count all discrete values of
> the differences file. Lower values mean a shorter loss of information.
> Another better scientific model would be calculating the cross
> correlation.
The problem with this is the way mp3 works: for a sine wave the difference
will be very small, for white noise it will be big because although the
frequency content of the encoded file is correct, the phase is messed up.
A single impulse will be encoded correctly, but two impulses close together
can be very wrong.
So the difference between the original and encoded *waveforms* is not useful.
If you want a technical measure of quality, you could look at the short-term
spectrum, as that is the main thing mp3 encoding is trying to preserve, but
to measure subjective quality you need something much more sophisticated (like
a team of listeners!).
Paul.
_____________________________
m a x i m | digital audio
http://mda-vst.com
_____________________________
This archive was generated by hypermail 2b28 : Sat Jun 08 2002 - 12:48:21 EEST