[vorbis] RC2 worse than RC1 and Beta4

[Monty]

On Fri, Aug 17, 2001 at 06:46:32AM -0400, M.r. Rage wrote:
>  >That methodology is flawed. You will see an increase in noise in a
>  >ramp at about 11kHz, following the ATH, because the ear is less
>  >sensitive there. This does *not* imply a net increase in
>  >spectral energy.
> 
> 
> Fine. My example was never intended to be anything more than an 
> illustration of what I'm hearing. I've been a professional audio engineer 
> since 1992, so I think I have a pretty good idea about what I'm hearing.
> 
> Based on what I'm hearing in the upper sibilance range, I think there's 
> another bump at around 7k. (Still using b4 128k)

I'm not arguing you're not hearing it actually... just that the
proffered explanation is not correct.  The fact that you identify 7kHz
and 10kHz matches the code in fact; those two places are exactly
where the code is relaxing the resolution given to noise (the bands
above 7-8kHz, and again at the bands above ~11kHz).

rc2 *does* currently increase the real spectral noise energy at 11kHz
plus. This is due to point stereo not having the code to actually do
the correct trig when rotating phase.  It fudges it for speed and adds
up to a factor of sqrt(2) to each spectral line it rotates.  Many,
many picky people are hearing that and I'm going to do the rotation
correctly in rc3.

b4 may be giving a similar impression, but the perceived increase in
energy is false. What many people describe as an artificial crispness
in the b4 high end is actually due to the opposite reason than users
claim.  It is not 'crisping' noise in the high spectrum, it is
actually smearing it.  Rather than hearing exact harmonics above 10k
(for example, in voice where such harmonics are important), one hears
the harmonics somewhat fuzzed.  Because the harmonics are not clean,
they're more noticable (not exactly mathing lower harmonics or
fundamentals), and thus a user could think that's a boost.

Careful spectral analysis should show this explanantion to be correct.

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.