[vorbis] Beta3 impressions

[Monty]

> - In comparison to beta2, subtle high-frequency artifacts seem to be gone
> (though they were small in beta2). Good job there! :-)

This is mostly the change of a single (out of ~50) masking curve.  I'm
thinking seriously of collecting my own high frequency data as I
suspect that the 8kHz+ data collected by Ehmer and Fiedler have some
hidden caveats.

> - Velvet.wav also sounds better, but transparent quality is reached
> at -b256+.

Velvet.wav is hard for reasons I describe in a just preceeding email.
It's not that the encoder is unaware of the problem, just that it
currently is missing the mechanism required to deal with it. 

> - Horn.wav still sounds very sucky, mode -b256 gives ~100kbps (this
> is understandable because in this sample practically there are no sounds
> above 5kHz) but the artifacts can be easily heard. 

Oops, I may have screwed up here.... not horn23_2.wav?  I may need to
re-grab Horn.wav then. :-(

> Mode -b350 gives
> ~310kbps but still there is a difference (subtle though). I also noticed
> that you get similar artifacts when you encode simple tones (like 440Hz or
> 1kHz), no matter what bitrate is used.

The encoder can't currently fill in strong pure tones with 'very
surplus' bits right now.  The higher resolution modes just back off
(and then remove) the noise masking.  That doesn't affect encoding of
pure tones at all.

> 
> - The most occurring artifact in -b128,160,192 modes is pre-echo (-b256+
> seems to be free from this curse). 

b256+ simply eliminate noise based masking entirely, throwing bits at
the problem.  Vorbis currently does an exceptionally poor job of
distinguishing between correlated and uncorrelated noise, so it
doesn't know to throw bits at correlated noise that causes preecho.

> The leader in eliminating pre-echo is AAC
> (Liquid Audio, Low Profile AAC), because even at ~128kbps stereo there is
> very little or no pre-echo (Temporal Noise Shaping works very well
> obviously).

...and is totally unusable due to patent complications.

> In his interview to Slashdot, Monty said that wavelets will be used in
> Vorbis (after version 1.0). I guess that wavelets will be able to eradicate
> pre-echo more thoroughly ?

Yes, wavelets are my attempt at solving the problem correctly.

> - There is also an improvement (from beta2) with gapless tracks, but small
> clicks are still there (not often though). I guess the only 100% solution
> is, like Monty said, to add a feature to the encoder so that encoder knows
> that it should match the end of one file with the begging of another one.

Correct summary.

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.