>
> This is Wan-Chieh again. I downloaded the LAME software that patches
> GPSYCHO to dist10 and makes a modified MP3 encoder. I was impressed by the
> encoding speed and the encoding quality that "Fake" mode encoding achieves,
> and I noticed that there are still some bugs in "Joint" mode. What
> surprised me was that without MS_STEREO coding, there are a lot of
> improvement in "stereo" mode over the ISO original encoder. What are the
> major differences between your "stereo" mode and the stereo mode of ISO?
>
Thanks! it's always nice to here other people say it sounds better
since I'm probably biased. So are you saying the Fake ms_stereo
sounds better than regular jstereo? I would be interested in
any samples which clearly show this.
Yes, I would say that the improvements in quality are not due primarily to
ms_stereo, but to the following: (more details are on the web site,
www.sulaco.org/mp3/gpsycho/gpsycho.html)
35% l3pcy.c improvements: The data being analyzied by the psy
model is not in sync with the data being encoded, plus none of the FFTs
are correctly centered over analysis windows. This is
why the ISO model switches to short_blocks 1 frame after
the sharp attack!
35% bit allocation: the outer_loop fixes that I mentioned in a previous
message. I have some questions about what to do here for ms_stereo,
(see the web page, and below).
20% bit reservoir control. tunning the PE formulas so that it is possible
to build up the bit reservoir helps a lot with pre-echo, (if pre-echo
can be detected)
10% ms_stereo. Like Weg mentioned, we dont really know how to do the
'sparsing' mentioned in the ISO docs, so I'm not really sure how to
get any savings out of this. I end up just allocating a bigger
percentage of the bits to the mid channel (tuned by looking at the FhG
mid/side bit allocations).
Mark
