>
> It's not sure that 1quantization error in band 0 and another in band 20 (ie
> 2 errors) are better than 1 error in band 9.
> But I think you'll agree that 1 error in band 0 or in band 20 is better than
> 1 error in band 8 or 9.
>
> The actual implementation will choose the last quantization with the over
> needed value. It means that one error in band 8 can be used instead of one
> in band 20 wich could have been found in a previous iteration.
>
> So I think that it might be a good idea to consider a "ear sensitivity"
> coefficient for each band to help in the choice when we got 2 quantizations
> with the right over value.
>
> This seems to be, to my mind, a good idea, but I perhaps missed something.
>
This sound good to me too. We could add something so that when
choosing between two quantizations with the same value of "over", it
computes some kind of weighted average of the distortion in each band.
If I get around to it, I will try and post the frequency ranges of all
the scalefactor bands (from the ISO standard). People with more
audio experience than me will need to say which bands are more
important than other bands!
Mark
--
MP3 ENCODER mailing list ( http://geek.rcc.se/mp3encoder/ )
