[vorbis] Vorbis determined to be as good as MPC at 128 kbps!

[Monty]

On Tue, May 25, 2004 at 09:02:50PM -0500, noprivacy at earthlink.net wrote:
> From: "Monty" <xiphmont at xiph.org>
> 
> 
> > We're lagging because there's other work to be done now.  However, the
> > reference implementation is *not* intended to be low quality.  At each
> > release, it tends to be the highest quality general purpose encoder.
> > here simply hasn't been a tuning release for a while, and so others'
> > tunings have surpassed the reference.
> 
> At some bit rates, probably.  But at the higher quality settings that most
> people used, even Garf's old version is still better than the newer official
> version.

Yes, because of impulse tunings.  I did mention that.

> > Not true; at each tuning release, I *did* integrate Garf's changes,
> > either in concept or whole-cloth.  Garf simply hasn't released new
> > tunings for a long time.  The one Garf change that was never
> 
> Well.... Let's just say that Garf seems to have a different opinion of that.
> And that his opinion is strong enough that he's abandoned Vorbis completely.

Having talked to him about it, he said "I don't have time to work on
Vorbis because I got a good paying job working on AAC, and I got it
thanks partially to the work I demonstrated on Vorbis."  Either you're
fixating on a moment of frustration, or Garf lied to me.  He doesn't
seem like the kiss-up sort, so I'll assume he wasn't lying.

*shrug*

That said... I don't generally fixate on squeezing out incremental
improvements by tweaking arbitrary numbers in the codec.  Arbitrary
numbers are the enemy of a good model, and I concetrate on eliminating
them and improving encoding results by making changes to the psych
model that produce predictable, explanable, non-arbitrary results that
don't need to be arrived at via exhaustive trial-and-error.

That's not to say there aren't large improvements left to be made with
the psych model already exists exactly by finding improved arbitrary
numbers, or other short-term engine fixes; you can get quite alot out
of tweaking the Vorbis what's there now.  Other encoders have been
demonstrating that in spades.  I, personally, am *very* thankful for
these parallel tunings.

But... folks seem to be reading alot into appearences when what we
really have is a difference in development strategy.  I think this
approach is relatively healthy... dev groups are out making solid
short-term improvements and getting real results *and* that doesn't
distract me from the Grand Strategy.  Don't pish the Grand Strategy.
Without it, you wouldn't have Vorbis at all.

Another case of mistaken appearences: My priority has not been to drop
my direction to integrate tunings. The work I do, in fact, eventually
aims to render the tunings obsolete (Garf did express frustration with
this at one point; he'd occasionally be ready with a set of tweaks
just in time for a new encoder infrastructure generation to render it
all irrelevant).  It can be frustrating, but it is all in the name of
building a better codec and for no other reason.

The HF boost is a modern example.  Sebastian's paper correctly
identifies one major source of the slight HF boost people noticed in
Vorbis some time ago. The solution he suggests is also perfectly
workable.  However, it can also be corrected more flexibly by
extending noise normalization to properly correct for all energy
gains/losses during quantization, and that is the way I plan to handle
it in the next major reference encoder release.  However, that release
is still some time off (as OggFile is my current priority), so it's
*perfectly reasonable* to use Sebastian's improvement now... however,
folding it into reference immediately means that I have to rip my
ongoing work out and put in Sebastien's for a release, then for the
next release rip Sebastian's approach back out and put mine back in to
get back to the long-term approach.
 
...and from a simple technical basis, people come up with all sorts of
conspiracy theories.  Conspiracy theories make Baby Jesus cry.

Monty
--- >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.