[vorbis] new ogg stream

[Kenneth Arnold]

On Wed, Sep 05, 2001 at 09:47:39PM -0400, Erik Kruus wrote:
> The stuff I did last year poking holes in the vq lattices makes me
> *guess* you might be able to peel around ~20% before things got
> unbearable.  (One may do better, they are different in what 'bits'
> get 'thrown away').  When you talk about halving the bitrate, you probably
> need a retuned psy model, and to hit 32k, I think it will require
> post-1.0 work on the core algorithms... Unless the peelable stream is
> generated by only adding bits to the low-bitrate stream.  But vorbis'
> peeling is more of an "incidentally-it-might-sound-OK" type rather than
> "constructed-to-sound-reasonable" type as far as I understand.

Your explanation makes perfect sense and answers some questions about
the peeling structure that I was wondering about; I was actually more
interested in halving the bitrate, going mono (getting hopefully
around 32k or less), and dynamically peeling to accomodate the
streamer's buffers so as to maximize bandwidth usefulness (modem
connections are often spotty for streaming rate and full-rate
streaming doesn't allow any other Internet activity at the same time),
and after all that see how it sounds, comparing to the current king of
modem streaming, RealAudio. I have some ideas of my own to pull
bitrate down even more, but they would almost certainly require
bitstream extensions and might well turn out to have too little
overall effect to be worthwhile, so I'll report back when I've
analyzed the compression process and intermediate data some more
(Vorbis is complex to one not experienced in signal processing,
especially when trying to get the whole picture. I hope to assemble
the results of my experimentation into a Vorbis encoding step-by-step
diagram showing intermediate results and what each step is doing,
basically like vorbis.on2.com (which is immensely useful btw) but with
more focus on the algorithm and the ability to give the encoder an
arbitrary sample and have it do each step of the encode and show the
results graphically; what would be really cool would be to make this
into an animated Java applet or MNG that would show the PCM data being
read in, the windowing, transforms, blocking, and final stream
output).

(that was a really long parenthetical note; I guess I need sleep :) )


-- 
Kenneth Arnold <ken at arnoldnet.net> / kcarnold / Linux user #180115
http://arnoldnet.net/~kcarnold/



<HR NOSHADE>
<UL>
<LI>application/pgp-signature attachment: stored
</UL>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: part
Type: application/octet-stream
Size: 233 bytes
Desc: not available
Url : http://lists.xiph.org/pipermail/vorbis/attachments/20010905/aa80b46a/part-0001.obj