[vorbis] [new?] Streaming technique

[Merijn Vogel]

One of the problems is that ogg is an vbr-codec, the targets of the encoder
are `approximately 128kbit' and `approx 256kbit', not constant bitrates.

The coder could make different decisions about the blocksizes, the bitrate
per block, etc. The blocks have a bit of overlap, I think it would be rather
hard to do, and it could be easier to peel or to distribute different
streams..

just my 2 pennies,

-- Merijn

how soon is Real Soon Now ? 

On Wed, Jun 06, 2001 at 12:46:37PM +0200, Alex Iribarren wrote:
> Before I start, I just want to say that I know very little about audio
> compression or signal analysis. And I have no clue what a Lorentzian 
> is, but
> I guess it must be some way of mesuring a signal. How is nonlinear
> compression done?
> 
> If I have understood correctly both of Dr. Joerg Bergmann's messages,
> the
> technique I have suggested is impossible because there is no redundancy 
> in a
> 128kbps stream and a 64kbps one, meaning that you can't extract the 
> extra
> data from the 128kbps stream to reduce it to 64kbps. Is this actually 
> true?
> 
> However, if that is not possible, perhaps you can overcome the nonlinearity
> problem by splitting the streams before actually compressing them
> nonlinearly. When reducing redundancies in the sound file to encode, and
> when applying the perceptual filters, you could apply them in different
> degrees to both streams (lots of filtering for the 64kbps stream,
> and half
> the filtering for the 128kbps stream). After that, you could "substract" 
> the
> 64kb stream from the 128kb one to obtain the 64kb of sound that was 
> filtered
> from the original 64kb stream. Then you would actually do all the 
> rest of
> the compression.
> 
> 
> >
> > But such thinking is invalid for nonlinear mechanisms.
> > I will give an explanation from my science:
> > Given any curve, I may fit it with a set of 5 Lorentzians
> > or, for more accuracy, with 10 Lorentzians. Doing so, the
> > resulting 10 Lorentzians _wont_ include the less accurate
> > 5 Lorentzians as a subset! Of course, you may construct
> > such a fit, but it will waste accuracy (or, in ogg:
> > bandwidth for the same quality).
> >
> > J"org Bergmann
> > email at jbergmann.de
> 
> 
> As to what Kristoff Bonne <kristoff.bonne at skypro.be> said, I think that
> would simply be a matter of syncronizing the streams through buffering,
> not
> much more difficult that what is actually done now with webcasts.
> 
> >Two remarks (to continue this brainstorming-session):
> >- One of the things you should take care of; is that both 'streams' 
> should
> >have the same bitrate (or a bitrate that does not vary to much vis-
> a-vis
> >eachother).
> >
> >What I mean is this:
> >If the 2nd session that is started is an (additional) TCP-session;
> you'll
> >have to deal with the 'slow start'; having a quite varied bitrate 
> at the
> >start of the session.
> 
> 
> 
> 
> 
> 
> 
> 
> --- >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.


-- 
Uit de laatste TOM (time-wasters online magazine):
Nog leuker was wel het verhaal dat Leo van TUNIX in een sexshop in Amsterdam
met een stalen gezicht zo'n paar handboeien uit een rek haalde en vroeg "mag
ik er hier 700 van?" :-)

--- >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.