[vorbis] Meaningful encoder testing

[Aleksandar Dovnikovic]

Craig Duncan <duncan at nycap.rr.com> wrote:

> I'm very _glad_ you're a perfectionist!  With respect to particular
> "bugs" being revealed by certain test wav files, i'm curious as to how
> this works.  Does Monty have a "suite" of files that stress certain
> aspects of the encoding?  Are they developed on the fly?  Do they
> produce discernable differences (to a listener) or just "noise" as
> recognized in a waveform analysis?

Vast majority of test wav files are short clips from real music that
are hard to encode. These clips reveal, more or less, flaws that most
encoder have (again more or less) like pre-echo, hi-frequency
artifacts, joint-stereo bugs, VBR misses... For instance, LAME
has pretty crappy pre-echo handling, so pre-echo test files
like castanets.wav produce noticeable artifacts even at 320kbps.
There is no need for waveform analysis - you can hear the
artifacts easily.
Another type of test files are computer generated wav files, like
sine waves, sine sweeps, white noise... - tuning the encoder
to accurately reproduce these test files might seem a bit stupid
(who normally listens to something like that?) but it is a good
thing to know that psycho-acoustic model works on all kind
of (test) files - that underlines its quality.

Greetings,
Aleksandar

--- >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.