>
> I remember seeing somewhere that a big change in the needed bits ammount was
> sometimes an indicator for window switching.
> Perhaps anyone got an idea to use it?
>
I think this is the algorithm used by the ISO model that we want
to improve. I've seen in papers the claim that the number of bits
needed is a linear function of the pe (perceptial entropy).
So a surge in pe (pe>1800) is what the original code uses to
indicate window switching.
This misses some critical pre-echo events in testsample2.wav.
It might be because the ISO formula for pe is not very good?
I think the new AWS stuff we have in the code does a very good job
at window switching. I'm going to post something about that next.
Mark
--
MP3 ENCODER mailing list
