[vorbis] Models used in mp3 and Ogg Vorbis?

[Segher Boessenkool]

Jack Moffitt wrote:
> It has other fancy tricks up its sleeve as well.  It also has a large
> amount of basic improvements over ways of doing things in MP3 (example:
> MP3 splits data into subbands with a lossy transform, them does more
> transforms on those.  Vorbis just does one big lossless transform).

Mathematically, this is correct.  With computer arithmetic, though,
and especially with careless mdct implementations, the MPEG Audio
subbanding is a lot more precise ("less lossy") than the Vorbis mdct,
unless you take a _lot_ of care to make the mdct better (the easiest
approach is to base the mdct on an fft that has already been tuned
for accuracy).  All this is pretty much uninteresting, though, because
you can't hear the difference anyway (MPEG subbanding has error
smaller than -130dB or so, while the libvorbis mdct has about -60dB..-80dB.
These errors also tend to correlate with the audio data, so you can't
hear it).

> While Vorbis probably has many improvements
> in this model over other codecs (although I suppose it's possible for
> those codecs to borrow these back from us), the real hard work comes in
> representing the information in the most compact way possible.

I somewhat disagree -- further improvements removing more data
( == improving the perceptual model) will eventually give much
better gains than representing/encoding the data in a better way
will give -- in this last category, there's probably not room for
better than a factor of two improvement.  It's just that improvements
in the first category are much harder than in the second category.

> In essence, MP3 is quite old technology-wise.  Vorbis is fairly young.

That is a very important point.  MP3 is (one of) the first of a
generation.  We don't drive T-Ford's anymore, either.

> We have solved the same problem with novel and modern methods.

Well... some things in Vorbis are really new, for the rest, mostly
not-all-that-new methods are used -- which isn't a bad thing.

<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.