[vorbis] Where's the tool to do .ogg bitrate "peeling"?

[Segher Boessenkool]

> Funny you should ask.  Segher wrote a "stupid" peeler in about 30 minutes that
> peeled any file down to 80 kbps earlier in the week.  I didn't listen to the
> quality, but was reputed to be fairly bad.  The peeler doesn't think too hard
> about which data to throw away, so the quality suffers a great deal.  A
> smarter peeler would be able to take the actual packet contents into
> consideration when deciding how much data to throw away in each packet.
> 
> If Segher is reading this thread, I'll let him jump in here and explain
> further.  :)

My toy peeler only looks at the sizes of all packets, and then tries to be
as smart as possible how much to throw away from each packet (a Vorbis packet
with some bytes chopped away at the end is still a valid Vorbis packet).

The big problem with this approach is, only looking at the sizes of packets
does not really give enough information to make a good decision about which
packets should be chopped how much.

My next generation toy will do partial VQ decoding on the packets, so it
will have a much better change of deciding right.

> > Also, I vaguely recall a post that suggested that current RC3 files can
> > be peeled, but that there is much "peelable file creation optimization"
> > that can be done in the encoder to improve on the resulting peeled file
> > quality.  Am I crazy, or is that the case?  Even if I am off about that,
> 
> Yes, it is the case.  I've been told by Segher that the encoder could do
> more to make peelers happy.

It's just a question of ordering the bits.  More important bits (for the
audio quality) should be earlier in the packet than less important bits.
In that way, you're certain that less important bits get peeled before the
more important ones (within one packet).  rc3 is not too bad about this,
but could certainly be improved.

Anyway, I'll put the sources to the toy peeler up at http://www.xiph.org/segher
so more people than just #vorbis can try it.

<p>Segher

<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.