[vorbis-dev] Why LSP?

[Jean-Marc Valin]

> OK, so similar.  I need less distortion than that, but that's just an issue of
> throwing more bits.  50 bits is about 3% of a 128kbps stereo stream.
> 
> Have you looked at implementing the equivalent of floor0 using modified
> cepstrum?  It would be a reasonably easy thing to design as a drop in
> replacement and it would be interesting to see how it performs.  I don't see
> an obvious benefit to it (indeed I also see no immediate drawback either
> although I think LSP is still slightly less work and more stable), but I
> haven't looked at it in detail.

I'm not too sure I understand everything in floor0.c, but what I could do
though, is write an envelope_to_a_few_bits() function. I have a question though:
*if* I could write a function to perfectly (0 dB error) quantize the curve with
0 bits and in zero time (wow!), would it be that much of an improvement for
Vorbis?

In the same line of thought, a while ago, I've tried to modify vorbis to
quantize the MDCT coefficients (not the envelope) in the x^.3 domain (I have
some reason for thinking it may improve). However, didn't really find where to
apply the transformation without affecting the MDCT windowing/overlapping and
the masking curve. Any idea of where I should do the transformation?

        Jean-Marc Valin

P.S. I'm interested in the research part of vorbis, is there any area where
there's still a lot to be done?


-- 
Jean-Marc Valin
Universite de Sherbrooke - Genie Electrique
valj01 at gel.usherb.ca

--- >8 ----
List archives:  http://www.xiph.org/archives/
Ogg project homepage: http://www.xiph.org/ogg/