[xiph-cvs] cvs commit:

[Kenneth Arnold]

On Mon, Oct 22, 2001 at 01:21:26AM -0400, Monty wrote:
> > Woah. Sounds ... horrid. Seems locked to 128k still also (which yields
> > 133k on my test song). Sounds like a "last minute bug".
> 
> Send me the sample!

Checked again with your later commits, and now sounds much better
... that is, 128k Vorbis again! If you still want the bad sample from
the earlier commit, I can send it but I don't think you'll need it now.

> (yes, it's hard locked to 128 +/- a few percent; this is true locked
> ABR.  Some samples like fatboy will kinda suck).

Much better now though -- 127.5k.

> debug output.  What you sent looks normal. 

Just thought the huge negatives looked weird.

I'm looking forward to next release now :)

And I got MatLab from school and have a tentative source for a copy of
Mathematica so I can start experimenting with some ideas of my own --
do you have any interesting leads for me to follow? I was thinking of
some way to take more advantage of temporal similarity among samples
(in other words, code most pop music in a few hundred kB), but I'm not
sure where to start. How about a reality check ... how much does
Vorbis /not/ account for self-similarity (like how optimally do the
codebooks allow the spectral energy floor to be coded, and could this
be improved)? I've understood a good part of Vorbis (at least the
parts that aren't channel coupling and psycoacoustics), but don't have
any feel for how those elements actually work in a real audio sample
no matter how hard I've tried to get one.

Basically, what's beyond Vorbis? i.e. either more computationally
complex or just plain unstudied (I'd prefer the second if you have any
ideas).

thanks for everything,

Ken


<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/commits/attachments/20011022/4f04d671/part.obj