[vorbis-dev] Peeling Specification

[Segher Boessenkool]

Alan MacDonald wrote:
> 
> Segher Boessenkool wrote:
> > Alan MacDonald wrote:
> >>Does it make sense to actually
> >>use new codebooks, instead of just chopping off the finer values?
> >
> > That requires recoding, and is much more expensive compute-time wise.
> 
> More expensive than peeling, but not even close to the compute time
> needed for full encoding.  (i.e. no transforms or masking calculations)

It's cheaper than decoding, even (about half as expensive, roughly).

> >>Or
> >>are the codebooks structured in such a way that chopping a q10 stream
> >>down to around q5 size, would yield about the same results as recoding
> >>the residue values with the q5 codebooks?
> >
> >
> > Nope.  That's why peeling doesn't work too well.
> >
> 
> Ah. So peeling doesn't really live up to the hype?

No, I meant that the current libvorbis encoding modes don't generate
very peelable streams.  Whether it is possible to _do_ generate nicely
peelable streams without much overhead is an open question.  (My vote is
on: yes, it's possible, but it requires (much) bigger codebooks then we
use now).

> Although, I think
> any of the hype created has not been by xiph, but by others just
> speculating.  Maybe it might be worth approaching this from two
> directions.  Straight bit peeling for the streaming folks that can't
> afford the CPU hit, and the new codebook approach for archivist that
> like to encode at high quality, but would want lower quality versions
> for portable playing (handhelds, laptops, etc).
> 
> I don't know, just some thoughts, is that feasible? good idea? waste of
> time?

Let's say I hope I'm not wasting my time ;)

> I'll be away for the next week, but would definitely be interested in
> continuing this discussion when I get back.

<p>Cheers,

Segher

<p>--- >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 at xiph.org'
containing only the word 'unsubscribe' in the body.  No subject is needed.
Unsubscribe messages sent to the list will be ignored/filtered.