[vorbis-dev] Sample still has horrorible distortions... anyone know why?

[Monty]

Ah, OK, just rechecked these with graphing instead of only listening.

Sweep.wav freaked me out at first because I *have* that sample, and it
was perfect...  It turns out ours are different in that mine is
attenuated -2dB.

On sweep.wav your problem is clipping.  Vorbis is psychoacoustic only
and the psychoacoustic engine doesn't know where the 'rails' are.
You're feeding in a sample that's going full range all the time, and
Vorbis is replicating it to within .5-1 dB... which according to its
calculations is enough.  Unfortunately, that is causing clipping which
looks and sounds pretty bad.  Attenuate your sample slightly and
you'll see the problem go away.

Mechanisms for backfilling more resolution/attenuating the waveform in
this case won't be in place until beta 4.

Velvet.wav: This sample is hard because it requires short blocks where
there's a heavy bassline but little/no lower midrange masking.  In
short blocks, bin 0 represents roughly 0-350 Hz (almost five octaves!)
and thus can require an absurd level of precision where the DC offset
is large, but there's no midrange to mask frame boundaries.  Beta 3
does not yet have the mechanism to fill the required bits in.  (It
should still be *much* better than beta 2, but I know it's not yet
perfect).

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-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.