[vorbis-dev] another entropy coder that might be very useful

[Sebastian Gesemann]

Hello, Yann !

Thanks for showing interest in optimizing Vorbis.

<p>On Mon, 26 Apr 2004, Yann Guidon wrote:

> Hello,
> 
> I want to let you know about an algo that will soon be published
> and that may be of interest to Vorbis (and other Ogg projects) :
> 
> http://f-cpu.seul.org/whygee/ddj-3r/ddj-3r.tgz

(Note for other users: This archive does not contain a
subdirectory)

Correct me, if I'm wrong. As I understand your algorithm,
you suggest to sort all samples first, do some "folding" by adding
a certain value to all samples so that they are positive, compress
the sorted list of positive interegs with your RR algorithem and
store some side-info about the permutation so that the original
order can be restored.

I don't see why this should be superior (in terms of compression
efficiency and speed) to the entropy coding layer which is
currently used. I also doubt that the RR step for coding the sorted
list of positive integers is near optimal.

Perhaps I missed something. It would be nice of you if you can
enlighten us. Did you successfully used this sheme as replacement
for JPEG's entropy coding layer ? Or was it still theory ?

You might want to compare it to
http://www.xiph.org/ogg/vorbis/doc/vorbis-spec-res.html
It's a detailed describtion of how the residue vectors are coded.
After the "floor-curve-subtraction" and quantization step the
samples are mostly uncorrelated signed integers (*) with varying
variance (higher variance in the low-freq part with some amplitude-
peaks (tonals) and lower variance in higher frequency regions due
to higher quantization.

(* quantization and coding can be done simultaneously by VQ, but
currently it's like scalar quantization to integers and coding them
in groups with a specific set of codebooks)

> I have no time to look closely at the Vorbis source code,

You don't have to. I would suggest to inform yourself first about
how the residue coding is done in Vorbis by checking the link I
gave above.

<p>Greets,
Sebastian


--
PGP-Key-ID (long): 572B1778A4CA0707

--- >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.