[For unknown reasons, this message was rejected by the mailing list
yesterday. I'm resending it today. ]
>
> Hi, I'm from Germany and I've checked the paper out.
> In section 4 ("Control of the window adaption"), it reads:
> (please note that this is my translation from German)
>
Thanks Andre for the translation! I've added some code to compute
these quantities to next release of LAME. I didn't use the formulas exactly
as written, but took the idea of looking for rapid energy changes
by doing the following: In each granule, the psy-model has already
computed 3 short FFTs centered over each third of the granule.
I use this information to compute the 3 energies and see how much
variation they have.
I also added Gabriel's pe adaptive switching code to the next release.
This seems like a good idea because I've also noticed many times when
the FhG will use two short blocks in a row and LAME will only get the
first one. Gabriel's code (if I understand it) will make it easier to
trigger the second pre-echo detection?
I haven't tried any of these ideas out yet, and the code is
turned off by default (#ifdef AWS and ENER_AWS).
A description of a good pre-echo problem is now at
http://www.sulaco.org/mp3/quality/preecho.html. There's an example
(and exact frame numbers) where FhG switches to short blocks and LAME
doesn't, resulting in a noticible flaw. There is also a sample where
the pre-echo 'improvements' I've tried in the past will generate too
many short blocks.
Mark
--
MP3 ENCODER mailing list
