Vorbis - Jun 2002 - peeling as I understand it (was Re: When will quality increase be unnoticable?)

>> Is bit-peeling going to be real (or just a rumor forever)?
> Apparently the RC3 streams are capable of being bit peeled, however the
> tool to do so was looking likely to be quite complex. I believe the plan
> was to have RC4 produce streams that left better hints for the peeling
> tool, so as to make the tool simpler and faster, but I doubt we'll see
it
> until post-1.0. 
NOTE: this is my understanding of the situation.  I am not a Vorbis developer.

RC3 streams are peelable.  No tools exist to do so. (Except a 
proof-of-concept hack that (I think) Segher made at some point.  Or maybe 
Monty.)

RC4 is focused on tuning the lower sample rates (below 44.1 kHz) and on 
getting a proper bitsteam management engine in place.  This is great news for 
people doing streaming with Vorbis, since rc3 had noticeable quality 
improvements over rc2, but broke -M maximum-bitrate functionality.  RC4 
doesn't address peeling at all, AFAIK.

I believe that RC4, coupled with proper documentation (file formats, 
bitstream formats, API docs, etc), will become 1.0 relatively soon after 
RC4's release.

After 1.0 is out the door, then some attention will be paid to
"hinting" to
make peeling easier, and peeling tools will finally be written.

It is suspected that RC3 or RC4 or 1.0 files may not peel perfectly.  That 
is, if a file is encoded at (for example) q9, and subsequently
"peeled" to
q6, it may not sound quite as good as the same file originally encoded at q6. 
 So they'll peel, and the bitrate (and filesize) will be right where you
want
it, but the quality may be sub-optimal.  But this depends probably on how 
"smart" the peeler is.

It is also suspected that post-1.0 files, with their "hints" or
whatever the
heck is going on, will peel "perfectly".  That is, a file encoded at
q9 and
then later peeled to q6 will be identical* to the same file originally 
encoded at q6.

(* By "identical", I mean that the two Vorbis files will decode to the
same
WAV and not necessarily that they are byte-for-byte identical, though maybe 
they will be.  I just don't know enough about it to say for sure.)

So that's the situation as far as I know.  But remember, I'm not a
Vorbis
developer so I could be completely wrong.


-- 
Graham Mitchell - computer science teacher, Leander High School
"Compassion should never determine our beliefs about sin; it should 
only determine our response to those who struggle with it."
	-- Harry Schaumberg, "False Intimacy"

--- >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@xiph.org'
containing only the word 'unsubscribe' in the body.  No subject is
needed.
Unsubscribe messages sent to the list will be ignored/filtered.

> -----Original Message-----
> From: owner-vorbis@xiph.org [mailto:owner-vorbis@xiph.org]On Behalf Of
> Graham Mitchell
> Subject: peeling as I understand it (was Re: [vorbis] When will quality
> increase be unnoticable?)
> That is, if a file is encoded at (for example) q9, and subsequently
"peeled" to q6, it may not sound quite as good as the same file
originally
encoded at q6.
So they'll peel, and the bitrate (and filesize) will be right
where you want it, but the quality may be sub-optimal.  But this depends
probably on how "smart" the peeler is.

<p><p><p>Well I wish someone would give a clearer explaination
of what is meant by
peeling, because there are at least two ways, that I can see, to effectively
reduce the bit-rate of a vorbis file with out completely decoding then
re-encoding.  The simplest way is to drop some of the residue codes, which
would give you a sub-optimal result, as Graham mentioned.  The second is to
decode the data to right after the codebook stage and then use a new set of
codebooks (ones for a lower quality level, and re-encode) which should be,
in theory, quite effective at reproducing what a file encoded at the lower
quality rate would produce, I think...

Although this would be a little more CPU intensive, it is still much less
intensive than encoding on-the-fly because you never leave the frequency
domain.  But, I think what we are talking about is the first scenario, where
coarser values are used for the residue, because of the layered nature of
the residue quantization.  Does anyone know?

Thanks,
Alan

<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@xiph.org'
containing only the word 'unsubscribe' in the body.  No subject is
needed.
Unsubscribe messages sent to the list will be ignored/filtered.