[vorbis-dev] extremely noticeable artifact (britney-bug)

[Gian-Carlo Pascutto]

----- Original Message -----
From: "Monty" <xiphmont at xiph.org>
To: <vorbis-dev at xiph.org>
Sent: Sunday, August 26, 2001 7:51 AM
Subject: Re: [vorbis-dev] extremely noticeable artifact (britney-bug)

> Sorry, I unexpectedly took the vacation I needed by getting sucked up
> into the run crew of a local theatre troup, wherein I've immersed
> myself in the mercifully Net-free atmosphere of musical theatre.  I'm
> surfacing again and I'll be getting everything tested toward rc3.

Aah, nothing like coming back from meatspace and having a fresh pile
of bugs to hunt.

> I've not tried the sample yet, no, but I've grabbed it and I'll be
> checking it with the rest.  Also, I hear you have tweaked your
> super-mode some more, I'd like to have a look at the current version
> ;-)

Hmm, it doesn't have anything newer besides the stuff we discussed
on IRC:

I started with something halfway between 128 and 160kbps mode.

Changed

a) lossless coupling (until you fix the current issues)
b) lowpass at 19kHz
c) postecho threshold (fixes problems with some of Dibrom's nasties)
d) noise masking in short blocks (aka just use lotsa bits to prevent
preecho)
e) ATH curve (set to lineconservative, just psychological change)
f) noise masking above 8kHz (Prodigy problem sample)

Except for castanets, it seems to do an ok job. It's certainly
nonoptimal, I just stopped when it started to sound ok to me.

I had the pleasure of abusing an expensive stereo yesterday instead
of my SB128 and used the occasion to burn a CD with 3*5 songs, each
time original, LAME with newest --r3mix (VBR-MTRH), and Ogg with my
mode. I scrambled them in the CD burning app, so it was a blind test.
(I used spectrum analysis after the test was done to figure out
what was what)

The results were (lower is better):
1. Ogg        9 points
2. original  10 points
3. LAME      11 points

Basically this means that I had no clue which was which. Sometimes
there were audible differences but there wasn't anything that clearly
sounded worse. It would probably still be pickable when cutting out
samples and ABXing them, but what I'm really trying to say is that
I wouldn't hesitate any more to turn over my CD collection to either
of these modes. The nice thing is of course that the Oggs were
significantly smaller than the corresponding MP3's.

These results are very interesting from a marketing POV:

Ogg offers BETTER than CD quality audio!

Monty, I've got another question for you. Can you write up some
kind of 'roadmap' that shows what/when you are going to work on
in the coming months/year? Estimates that are years off are fine.

e.g.:

august/september: fix bugs, some quality issues
septemer: release rc3, fix bug, release rc4 that becomes 1.0
october-december: clean up mode infrastructure, make encoder tunable
                  via flags/settingsfile, make peeler
january: party
february: gather tuning done while you were partying, release 1.1
march-august: go into deep wizardry mode and develop wavelet support
september: merge stuff together
october: release 2.0rc1, rc2, rc3pre1, rc3pre2ac1
december: release 2.0, world domination, more partying
january: get deaf because of a last bug that slipped through

Well, you get the idea.


--
GCP

--- >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-dev-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.