[vorbis] further development

[Keith Wright]

> From: Robert Michel <Robert.Michel at post.rwth-aachen.de>

> And I talked with one Speex developer, Jean-Marc:
> >
> > I think in most cases Speex will take less time
> > encoding than MP3 and Vorbis.
> 
> Try out about CPU power for deconing.  I got the
> impression from the Speex homepage that speex has a
> more simple algorithm then other codecs, so it should
> be faster and have smaller bitrate for same quality.

I got the impression that Speex uses an O(n^2) algorithm,
(Levinson-Durbin for CELP), while Vorbis uses an O(n log n)
algorithm (some kind of twisted FFT), so I am dubious that
it will be faster*.  The algorithm that is simplest, in the
sense of easier to invent and explain, is often not the
faster or better one, which is discovered after decades
of research.

*NB: I don't say it isn't faster, just that a glance
at the home page is not sufficient proof.

On a fast enough processor, speed of decoding may have
no direct relation to latency (delay till sound begins).
Quality comparison between a music codec and a speech
codec is a tricky business.

I have not done comparitive tests, and it seems nobody
else has either.  Such tests are the only way to really
answer such questions, and the answers may change with
the next release, or on different hardware, or after
tuning for some particular purpose.

> Es gibt nicht Gutes, ausser mann tut es.
> (There is no good without someone does it.)

I know little German, but enough to know that sounds
far better before translation to English.  But if
you can figure out what it means, it is still true.


-- 
     -- Keith Wright  <kwright at free-comp-shop.com>

Programmer in Chief, Free Computer Shop <http://www.free-comp-shop.com>
         ---  Food, Shelter, Source code.  ---
--- >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-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.