[vorbis-dev] vorbis-utils features ;)

[Beni Cherniavsky]

On Fri, 23 Mar 2001, Gregory Maxwell wrote:

> On Fri, Mar 23, 2001 at 12:54:07PM +0200, Beni Cherniavsky wrote:
> Talked to Monty. Found out I was wrong: Vorbis already has a time-dampened
> peak value... I thought I was being clever, but Monty of course, he was
> already cleaver first. :)
> 
It's the documentation problem.  If the docs were up-to-date, everybody
would know how smart and powerfull vorbis really is but now they can only
guess/deduce it from the high quality they experience :)

> 'Lookahead' might not be totally stupid, after all if you've got it cranked
> so loud that the peak is 16db over your threshold of pain, you aren't really
> going to care if the super-quiet first 10 seconds where a little less then
> perfect. Unfortunately the fact that it would pretty much require a two pass
> encoder relegates this optimization to a future encoder which is already
> multipass for other reasons.
> 
It's not multipass, it's a fixed-size buffer.  Maybe it can be useful.
Alas, it adds latency, which is a Bad Thing for all encode-and-stream
applications.

BTW, how symetric is Vorbis with respect to time?  If I get some time i'll
experiment with encoding a wave reversed in time (and reversing the
decoding of course so that I can understand something) - will this give
quality close to normal?  Should it?

> > All this is assuming one won't do strong band-filtering to listen closely
> > to quiet frequencies of the sound but Vorbis assumes this already.  
> > People should be punished for certain quality-problem-hunting listening
> > habits ;-)
> 
> Doing that sort of thing would make sence... providing you where evaluating
> the quality of an audio system to be used by aliens.
> 
> :)
> 
Where can I get some measurment data for the cosmo-acoustic model?  Are
there any publications on it?  Exozoology might help...

__unrelated_rant__ {

This reminds me of a thought I had lately...  The brain is evidently using
percuptual time compression.  Time is streamed into your consiousness in a
VBR stream, using a sophisticated attention-based scheme resembling
Vorbis' short-block toggling.  The computational power required to process
this stream varies correspondingly.  The psycho-chronostic model is a
constantly adapting algorithm smartly implemented by feedback from the
complex that is processing of this stream.  The ordering of bits inside
this stream is by importance.  However apparently no rigid division into
blocks exists.  More important bits arrive faster.  Even more amazigly,
bit peeling is implemented directly by the medium storing this stream for
later replaying - it is done gradually with time, taking the modifications
to the psycho-chronostic model that have happened meanwhile into account.  
Actually the bit peeling proccess doesn't through the bits immediatly but
gradually migrates them deeper and deeper in the memory hierarchy, thus
retaining the low-bitrate versions immediately available, while allowing
to reconstruct the stream at any needed quality at the cost of access
speed (actually the quality is limited by the storage medium fault rate
that increases as you sink in the memory hierarchy).  As if all this
wasn't enough, the bitstream format is always compatible without being
frozen: as time passes the reconstruction of the same stream will become
richer.  This function is also partly performed by the storage medium.  In
other words we witness a compression rate that improves with storage time
both by lowering the bit rate and by raising the quality!

Vorbis has a lot to learn yet (: excusable by the huge development effort
that went into the above system, compared with Vorbis' :)

} /* __unrelated_rant__ */


-- 
Beni Cherniavsky <cben at crosswinds.net>
                 (also scben at t2,cben at tx in Technion)

10001110111100111100001001010 m/s

--- >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.