[vorbis] When will quality increase be unnoticable?

Alan MacDonald newslists at warpzillion.com
Fri Jun 21 08:51:47 PDT 2002



> > > -----Original Message-----
> > > From: owner-vorbis at xiph.org [mailto:owner-vorbis at xiph.org]On Behalf Of
> >
> > > I think 4.99 uses a 8-phase-stereo, all over 5 uses lossless-stereo
> > > (therefore the big bitrate jump).
> > >
> >
> > It's not 8 phase stereo its just a lossy coupling of left and
> right below
> > q5...All that documentation about the different phase stereo coupling is
> out
> > of date and not used in RC3.  If you want a further
> explaination on how it
> > works let me know, but it takes a little of vorbis background
> to explain.
> >
>
> Yes, that would be great!
> Thanks for correcting me.
>
> metrom

This is my interpretation on how coupling works in RC3.  When vorbis files
are encoded the two main elements you end up with are the floor and the
residue.  The floor is basically a set of line fits to frequency domain data
after that data has been sent through the psycho-acoustic model.  The floor
is subtracted from the frequency data and the remaining values are the
residue.

Now for the coupling there are cut-off frequencies, below which the lossy
coupling does not take place, somewhere around 4kHz I think, but it varies
based on quality level.  Lower cut-offs for lower quality.  For the lossy
coupling you end up with a floor for both the left and right, but only one
residue value.  This value is basically an average of the left and right
residue based on what their floors are.  So in essence, you end up with a
coarser representation of the channel separation.  I hope that makes sense,
let me know if you have any questions.

This is all based on my reading of the source, but I'm pretty sure that's
how it works, although, if anyone knows better please let me know.

<p>Later,
Alan

<p><p><p>--- >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.



More information about the Vorbis mailing list