[vorbis] Question, problems re: Ogg Vorbis ..

[Aleksandar Dovnikovic]

"Merijn Vogel" <merijnv at sci.kun.nl> wrote:f

> Good luck to the vorbis team, my ears seem to be bad to me, I still cannot
> hear the differences between beta1 and beta3 ;-). Therefore this request,
> does somebody know about a website that discusses things like pre-echo,
> ringing and stereo errors so I can learn what to listen to ?

First, one small warning/suggestion: don't do things like this. It will ruin
your music experience because you'll train yourself and your ears and
you'll hear artifacts much easier. I know, I've done that... :-))

Anyway, if you're still interested, you can visit a page on LAME's
(open-source mp3 encoder) site and there you'll find many test tracks
(with explanations) that most encoders have trouble encoding properly:
http://www.mp3dev.org/mp3/gpsycho/quality.html
(Hint: most troublesome tracks are castanets.wav, fatboy.wav, spahm.wav,
velvet.wav and applaud.wav)

Also, you may check this forum - it is primarily used for mp3 discussion but
lots of people there are also very experienced listeners:
http://pub41.ezboard.com/br3mix

Finally, check this message that David McIntyre (he works for QDesign)
wrote on artifacts some time ago on mp3.com message board - it is a
good starting point because it explains most common types of artifacts:

Artifacts
(by David McIntyre)

A definition of artifacts is very difficult without audio examples (and I
don't have time to make some) and again, different people hear different
things, but here are the most common:

chirps: high-frequency distortions added to the audio, will sound like short
chirps, whistles, sometimes can shwoosh up like a high-pitched glissando
(slide).

pre-echo: the killer artifact that everyone has trouble with and the whole
TNS structure in AAC was designed to handle but didn't work as well as
expected (there's sentence). To hear this find a track with very precise
sharp hits and no background noise (the best is a castanets track we use and
is part of the SQAM testing disc). When you code it the attacks will be
smeared - if you listen closely or look in Cool Edit you will see that there
is an attack before the real attack, thus "Pre-echo". This is a huge problem
for MP3. The reason no one has completely solved it is basic filter theory -
as you improve time resolution, you lose harmonic resolution and/or vice
versa. Codecs that have little pre-echo tend to sound less warm and more
brittle.

Imaging: (my worst, so my description may be bad) listen to the placement,
in the stereo field, of all the parts. Also, get a sense for how "big" the
stereo field seems. When you listen to compressed audio you will find that
parts get moved towards the center or pushed out to the extreme edges. Also,
you will find that smooth pans from side to side seem to get quantized into
steps instead of a smooth continuum. The other thing is a simple, general,
loss of "width".

Pumping: pumping is a general term that can apply to frequency, amplitude,
imaging, etc.. This simply means something, lets say left channel amplitude
is going up and down fast enough to be bothersome and unatural, but slow
enough to be really noticeable. Maybe 2-4 times per second. Amplitude
pumping is the most common type and has been solved in most codecs, but can
be a problem.

Harmonic content/depth: simply, does it sound shallow or brittle (PAC's
biggest problem), or does it sound warm and ful.

Noise: has noise (static, hum, etc.) been added or removed. Is it centered
on one particular event or frequency range.

Other time resolution problems: other than pre-echo the most noticeable is
the watery sound many codecs get, particularly on voice. This has to do with
insufficient time resolution on the rich harmonic content in the vocal
spectrum.

Phasing: to kill AAC, or the QDesign Music Codec, trickling water, or rain
on a sheet metal roof with other sounds in the background will be very
difficult. In particular, the definition of the water will be lost and it
will start to sound like frying oil and not water drops. This is due to
quantization of the phase information in the audio. With all these water
drop hits - a series of like sounds close together - you get a lot of
signals that are similar and slightly out of phase with one another adding
up to make a noise-like sound. Usually, a codec will screw up these delicate
phase differences and tend to make it sound more like white noise (I don't
understand the details, but I know this is THE KILLER SIGNAL for AAC).

--- >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.