[theora-dev] Experimental release of Ghost/CELT 0.0.1

[xiphmont at xiph.org]

On Dec 10, 2007 9:33 PM, Timothy B. Terriberry <tterribe at email.unc.edu> wrote:

> For audio that's less clear. Vorbis gets most of its compression from
> using VQ, which only really benefits if you have non-power-of-two
> codebook sizes, which would require re-training codebooks...

What would be useful to a vorbis-like compression would be the
self-training aspects thus eliminating the need to pack the entropy
model (codebooks) into the header.  That's a worthy goal in a number
of ways.  I would require a spec update, of course, and it would also
require keyframing.

Monty