From: Takehiro T. <tak...@in...> - 2002-12-13 16:50:03
|
Gabriel> >Modified Files: > quantize.c version.h >Log Message: >fixed the quant_comp problem (probably introduced by me) >so bump the alpha version number I have 2 questions about this fix. 1) In this fix, you removed "quantcomp_alt_type" related code in quantize.c. This changes the result of the VBR preset (--preset standard). "quantcomp_alt_type" is used when ATH value is small. Is this what you want to do ? Have you tested with/without it and got better quality ? 2) There are 2 more such "ATH adjust" usage when VBR preset. In quantize.c, it is used to changed if LAME uses "scalefactor scale" or not. And, in psymodel.c, it changes nsfix values. This is needed because the part of psymodel ignores ATH adjust. More precisely speaking, in psymodel.c, LAME uses ATH values somehow. But it uses ATH value itself and ignore ATH adjust. I think this is why the VBR preset need "quantcomp_alt_type" and so on. In my experimental branch, all the ATH values in psymodel is scaled by ATH adjust value and removed "quantcomp_alt_type" and so on. And I think this is OK at least for my listening test. Does anybody think this is wrong ? If everyone think this is Ok, I will back-port the patch into the head branch. -- Takehiro TOMINAGA // may the source be with you! |