[vorbis-dev] One codebook for all audiofiles?

Gregory Maxwell greg at linuxpower.cx
Tue May 29 09:54:08 PDT 2001



On Tue, May 29, 2001 at 07:27:59PM +0300, Juhana Sadeharju wrote:
> [ I'm not in the list because I didn't find a digested version; please
> move the lists to sourceforge.net, and we would have the digested version.
> I read the replies from the archive. ]
> 
> Hello.
> 
> Would it be possible to allow Vorbis use the same codebook for multiple
> files? I could keep a 650 MB codebook on CD-R and use that for all my
> audiofiles. If that is possible, how much the compression ratio or quality
> would increase? How the codebook should be generated?
> 
> Do you have the algorithm documentation available? The distribution
> and the webpage has no documents.
> 
> [ Slightly offtopic: ]
> Because I'm wondering which is better at high bitrates (1:4 compression
> ratio), Vorbis or LAME, I would like to know how to compare the original
> audio and the decoded audio? With LAME, a simple sample-by-sample comparison
> failed badly (the error was as high as 7 bits). With ratio 1:4 I aim at
> archival quality and thus the overall error counts, not psychoacoustics.

At 4:1 you are probably looking at the wrong compressor: None of the
psychoacoustic transform coders are designed to work very well by the metric
you've set. Not that they won't sound good..

Shorten has a lossy mode that would achieve what you want, but I don't think
it's a really ideal coder..

The best coder for such a task would be a state-of-the-art lossless coder
(not shorten, perhaps once squish continues), coupled with a perceptually
shaping lossy optimizer, which I haven't seen any codec do.

I suspect that we don't have many 'slightly lossy coders out there, because
from a perceptual perceptive they would offer very little (no?) advantage
over fully lossy codecs (like Vorbis) operating at high bitrates for any
application I can think of.

Really, the *ONLY* advantage of a lossless codec over a good high-bitrate
perceptual codec (mp3 doesn't count mostly because of it's length quantization)
is generation loss.  If generation loss is not a factor, then sufficiently high
bitrate Vorbis should meet your needs (if it doesn't, let the list know, it
should!) with much more efficiency, if generation loss IS a factor, then no
loss at all is acceptable.

--- >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.



More information about the Vorbis-dev mailing list