I'm a bit worried about LAME's short block detection.
Consider the following sequence of short block
energies with a big surge between b3 and b4:
b1 b2 b3 b4 b5 b6
<100><100><100><2000><2000><2000>
If the frame boundaries are like
f1 f2
[b1,b2,b3][b4,b5,b6]
then LAME will not switch to short blocks, because
there are no big surges in f1 or f2.
But if it's a little bit shifted and looks like
f1' f2'
[b0,b1,b2][b3,b4,b5]
or
f1" f2"
[b2,b3,b4][b5,b6,b7]
then LAME will switch to short blocks, because of
big surges in f2' and f1".
Maybe I missed some point and it would be
OK not to use short blocks in frames f1 and f2 ?
Robert
--
MP3 ENCODER mailing list ( http://geek.rcc.se/mp3encoder/ )