On Mon, 6 Sep 1999, Robert Hegemann wrote:
> Hi to all,
>
> does anybody know why we calculate s3_s[][] in L3para_read
> and not using it further? Should we use it for the threshold
> computation for short blocks?
>
> > for ( b = 0; b < numCBANDS_s; b++ )
> > for ( k = 0; k < num_part_s; k++ )
> > ecb[b] += s3_s[b][k] * eb[k]; // s3_l replaced by s3_s
>
> Robert
>
That looks like another bug. s3 is the spreading function, which
has a simple shape as a function of 'barks', but the scalefactor
bands have different bark values (bval_s, bval_l) depending on
short or long blocks, so I think we should be using s3_s.
The ISO code was also using norm_l
instead of norm_s in the short block computation, but that
was fixed a while ago.
I'll fix this and add it to 3.26. It will create more energy
spreading between scalefactor bands for the short blocks, but
how this will affect the masking?
Mark
--
MP3 ENCODER mailing list ( http://geek.rcc.se/mp3encoder/ )
