Segher Boessenkool
2000-Dec-22 06:43 UTC
[vorbis-dev] Different floor, quality improvement
Hello all, Please try this "patch". It changes the way the noise floor is used for quantization in a not-so-subtle way. At the very end of _vp_compute_mask, add the lines: for(i=0;i<n;i++) flr[i]=.01f*sqrt(flr[i]); The .01 is there to ensure the current codebooks will work. We will really need different, newly-trained codebooks with this change; then the .01 can probably go away, though it won't hurt (I think). Be warned that the bitstreams increase roughly 50% in size with this change, so you should compare 128kbps new with 192kbps old etc. Please try this on tracks you find difficult to encode, especially very tonal stuff, I'll explain later what this really does, it's a bunch of math so I'll make a TeX file of it, Cheers && happy coding, Segher --- >8 ---- List archives: http://www.xiph.org/archives/ Ogg project homepage: http://www.xiph.org/ogg/ To unsubscribe from this list, send a message to 'vorbis-dev-request@xiph.org' containing only the word 'unsubscribe' in the body. No subject is needed. Unsubscribe messages sent to the list will be ignored/filtered.
Segher Boessenkool
2000-Dec-23 15:42 UTC
[vorbis-dev] Different floor, quality improvement
Some additional notes... Has anyone tried it yet? Segher Boessenkool wrote:> > Hello all, > > Please try this "patch". It changes the way the noise floor is used > for quantization in a not-so-subtle way. > > At the very end of _vp_compute_mask, add the lines: > > for(i=0;i<n;i++) > flr[i]=.01f*sqrt(flr[i]); > > The .01 is there to ensure the current codebooks will work. We will > really need different, newly-trained codebooks with this change; then > the .01 can probably go away, though it won't hurt (I think). > > Be warned that the bitstreams increase roughly 50% in size with this > change, so you should compare 128kbps new with 192kbps old etc.Using .028 instead of .01, you will get bitstreams of the designed bitrate.> Please try this on tracks you find difficult to encode, especially > very tonal stuff, > > I'll explain later what this really does, it's a bunch of math so > I'll make a TeX file of it,Some really inexact and quick answer: It makes it easier on the vector quantizer to make a "fair" decision. Now define "fair" ;-) (In this case: "statistically fair". Now define "statistically fair"... ) Merry Christmas to all of you, see you guys in the next millenium, Segher --- >8 ---- List archives: http://www.xiph.org/archives/ Ogg project homepage: http://www.xiph.org/ogg/ To unsubscribe from this list, send a message to 'vorbis-dev-request@xiph.org' containing only the word 'unsubscribe' in the body. No subject is needed. Unsubscribe messages sent to the list will be ignored/filtered.