[vorbis] masking problems in Vorbis beta?

[Monty]

> Hi all,
> 
> I just came across the following page: http://r3mix.net
> In the "News" section there is a fairly negative critique of Vorbis;
> especially it is mentioned that Vorbis "has terrible masking
> problems".

The test involved is not having masking problems, it's hitting the short-block
bug.  We are aware of this bug (and have been since before the beta); I'll be
sending out a complete description of the bug here in a few minutes. since this
is getting passed around alot.

It's just a bug, and not a very serious one (although it *does* have a big 
effect :-)  The guy who runs r3mix.net says he's happy to run new tests when I 
have a fixed version to send him.

Summary: log sweeps (and other artificial test samples with strong, isolated 
tones) trigger a bug in the encoder that causes it to choose almost all short 
blocks.  Short blocks are exactly the wrong choice for strong tones (as 
they're designed to represent impulses).  The result is terrible fidelity.

For comparison purposes, listen the following log sweep sample with short 
blocks disabled:

http://www.xiph.org/ogg/vorbis/test.ogg

...and you'll see that the sample is near perfect, 0-22050Hz, at less than 
25kbps/ch.

> There is also a Vorbis-encoded frequency sweep
> which shows strong deficiencies at high frequencies, but
> I suppose this is due to the quite low bitrate of the
> distributed encoder.

Actually, the beta went out with a 16kHz cutoff enabled.  There will be no 
cutoff in the next release.

> I would be glad if one of the Vorbis insiders could comment
> on that critique. Does the guy have a point or are the problems
> caused by the low bitrate?

They are problems caused by analysis bugs, and ones that will be fixed ASAP.

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.