No, I don't think so. Was: [vorbis] Vorbis 1.0 binary - Beta4 still the best?

Monty xiphmont at xiph.org
Fri Jul 12 11:22:41 PDT 2002



On Fri, Jul 12, 2002 at 01:59:02PM +0200, Zvezdan Dimitrijevic wrote:
> I am sorry to say, but after extensive tests I think that Beta4 -b 112
> kb/s are better than new v1.0 with -q 5 or -b 160.
> 
> Saddly but as I wrote before, I came to same conclusion with previous
> releases, RC2 and RC3.

Zvezdan... forgive me for being blunt, but you're on crack.  b4
attenuated high frequencies *much* more strongly than rc4/1.0.  For
one thing, everything was lowpassed by masking/ATH several kHz lower
in b4 (masking curves started rolloff at 16kHz), and the noise
pumping/phlanging was almost uncontrolled.

You're either doing something wrong or you just like the specific
coloring of beta4, akin to people who will never give up a trusty tube
amp, even if by every measure it's inferior to transistor, because the
like that specific distortion character.

B4's artifacts were much stronger.  A rundown: 

Perhaps you're mistaking lots of high frequency artifacts for it
making the high end 'richer'?. B4, compared to 1.0, has about 20dB
less tone differentiation (the maximum tone resolution in b4 was 22dB
for the strongest sounds, and about 12-14 dB for anything under that.
In 1.0, it's about 40dB for all tones) and about 6dB less 'depth' in
noise at -q 3.  Cymbals are much more stable and less watery/metallic
in 1.0 as a direct result of tuning out flanging (which some people
seem to like; it gives the cymbals more 'character' and can bring a
flat, boring cymbal more 'splash' or 'spin').

Although 1.0 does use lossy channel coupling, beta 4 was still using
floor 0, since deprecated.  It is deprecated exactly because it
*damaged stereo image fidelity* and resulted in unstable localization,
often making the stage spread artificially.  The LSP continuous
approximation in beta 4 caused sounds to wander and flutter block to
block; it was impossible to keep the image firmly planted in any part
of the spectrum because a small change, anywhere in the spectrum,
shifted the entire approximation, causing framing differences.

B4 also used a larger block size in mode AA (the old 112kbps mode),
which would cause much more noticable preecho on all attacks (eg, a
snare brush/hit, a ride cymbal tap).

All the above describes my utter disbelief that any aspect of b4 was
superior, subjectively or objectively-- unless, of course, you've
simply gotten very used to beta4.  The majority of the 'bug reports' I
get with every new Vorbis release are not really 'X sounds worse' but
'X sounds different'.  Extensive blind testing performed by myself and
third parties support all of my above assertions.

This is not meant as an attack.  it's meant as 'the world as well as
my own ears believe what you say is patently false.  We have evidence
that supports us.  Please spend some time justifying or explaining
your position carefully.'  Extraordinary claims require extraordinary
evidence.  Of the several thousand individuals who have personally
reported an opinion, you're the only one who's remarked that b4 was
superior.

(FWIW, there will indeed be the ability to force lossless coupling in
any -q mode as a post-1.0 feature.  I hesitate to do this, because it
will give alot of clueless newbies who heard information from some
friend that 'joint stereo wrecks mp3 quality' a tool to damage their
vorbis collections by pumping useless bits into stereo that should be
used for resolution and depth elsewhere.  Still, it may be useful to
some experienced users in some circumstances).

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.



More information about the Vorbis mailing list