[vorbis] Ogg/Vorbis Downsampling?

[Gregory Maxwell]

On Wed, Mar 21, 2001 at 12:15:02PM -0500, craig duncan wrote:
> > Yes, you could do it with the files you produce today. But, the results
> > would be sub-optimal because the encoder doesn't expect you to be doing such
> > a thing. In the future the encoder will produce output which is a bit
> > differnt (cascaded codebooks) so it can express to the bitrate peeling
> > utility what information it can throw away first.
> 
> Ah, thanks.  Do you mean to say that (currently), as you move closer to the
> end of each packet, the data in no way "correlates" with just the info
> associated with higher bit rates?  (So you _could_ peel it now, but the result
> would be random in terms of what was saved and what was thrown away?)  It
> wouldn't be accurate to describe this as "sub-optimal", though. :)  "express
> to the bitrate peeling utility what info it can throw away first" seems to
> imply that the "algorithm" isn't as i understood it, e.g. throw away however
> much you want starting from the end of the packet.  Maybe you couldn't throw
> away an arbitrary amount but, rather, the amount that corresponded to what was
> encoded in a pass across the frequency spectrum, which is a minor wrinkle.  Is
> that what you mean, or is my idea about this misguided?

Since we currently make only one pass across the spectrum you could peel by 
1) Zeroing out partitions (effectivly doing bandpass filtering)
2) Replacing codewords with shorter ones. (not likely much benifit for the
   quality loss)

When cascading is in use you can do the above as well seletivly throw out
the second 'pass' on varrious partitions thereby smoothly degrading the
quality rather the replacing words with wrong data or silencing partitions.

Since the second pass will be stored second, the throw away the end
technique will work as you describe.

--- >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.