>
> Hi All!
>
> there are 2 bugs in psymodel.c.
>
> Uum, this may be buffer overrun...
> we need
> if (chn<2)
> before setting uselongblock[chn], I think.
>
thanks! i just fixed this.
> and about this, there's more complicated problem.
>
> psymodel.c detemines block_type from the LR threshold,
> even when it uses MS threshold. I think this is strange.
>
> does anybody have any idea about it ?
> ---
It was done intentionally, and I think it is okay:
If an attack shows up only in the side channel, then the L & R
channels are very different and hopefully LAME would not be using
MS stereo.
If an attack shows up only in the Mid channel, then it would
also have to be in the L & R channels. The only time I can
see when it would not be detected in either L or R would be if
somehow (L+R) strengthens the attack and weakens the tonal components,
but I doubt this happens to often? To have the tonal components
cancel in L+R, the would have to be almost the same but 180 degrees
out of phase.
Mark
--
MP3 ENCODER mailing list ( http://geek.rcc.se/mp3encoder/ )
