OK. I'll have some more experiments about this.
Mark> The reason you cannot use maxnoise is that the output of a lapped
Mark> transform (like the MDCT) with very short transform lenghts (576 and
Mark> 192 in our case), individual frequency information CANNOT BE TRUSTED!
Mark> You have to do some kind of smoothing/convolution before this
Mark> information is meaninfull. For example, two identical signals, just
Mark> phase shifted and with different low-level noise, can generate very
Mark> different maximum values, but to treat these signals differenty would
Mark> be a mistake.
Mark>
Mark> This is the real reason why the psycho-acoustic model does not do
Mark> spectral line-by-line calculations, but first convolves down
Mark> to 64 "partition" bands. (They did not use partition bands
Mark> just to save CPU cycles!)
Mark>
Mark> This information is then averaged (or in the case of AAC,
Mark> they use min & max) down to the 21 scalefactor bands.
Mark> But what do you think of this variation on your idea? :
Mark>
Mark> 1. compute psychoacoustics in the 64 partition bands
Mark> 2. compute noise in the same 64 partition bands
Mark> 3. adjust the scalefactors for scalefactor band N so that
Mark> noise < masking, for all partition bands contained in band N.
Mark>
Mark> I think this is a good compromise between avenoise and maxnoise,
Mark> and actually simplifies the psycho acoustic model since
Mark> we dont have to map from partition bands down to scalefactor bands.
Mark>
Mark> Mark
--
Naoki Shibata e-mail: [EMAIL PROTECTED]
--
MP3 ENCODER mailing list ( http://geek.rcc.se/mp3encoder/ )
