[FFmpeg-soc] [soc]: r3561 - mlp/mlpenc.c

[Jai Menon]

Hi,

On Saturday 23 Aug 2008 9:14:30 pm Ramiro Polla wrote:
> ramiro wrote:
> > Author: ramiro
> > Date: Sat Aug 23 17:39:25 2008
> > New Revision: 3561
> >
> > Log:
> > Simple stereo decorrelation (mostly copied from alacenc.c).
>
> This is 1.4% worse than the previous fixed L, L-R decorrelation. There
> is some overhead because of changing coeffs, but it shouldn't be this
> bad. Any ideas why? (maybe estimate_stereo_mode() isn't that precise?)
>

Maybe 2nd order residuals aren't a good estimate. Try seeing if 1st order is 
better.

Regards,

Jai Menon