[Audacity-quality] [Bug 553] Zero crossings selection when short selections or white noise

[James C. <cr...@in...>]

> Try using find zero crossings on white noise.
>
> The results are poor, in particular:
> 1: A zero selection may result.
> 2: A good selection may be made obviously worse

I believe the problem(s) reported in this bug are fixed.

1. Does not happen any more
2. I've not seen happen.

The bug is largely dubious as a test case anyway.  White noise can be 
cut and pasted as you like without issue.  Zero crossings don't make 
editing white noise any better.  You're not going to hear new clicking 
that was not there before if you don't look for zero crossings.  It's 
the same reason that you can cut and paste into fricatives and sibilants 
relatively safely, but not so for vowel sounds.  It's harmonic sounds 
where zero crossings are relevant.

White noise is a dopey test case for zero crossings.  A better test is a 
mix of equal parts white noise and 4KHz tone.  Zero crossings does 
superbly well with the new algorithm there, much better than the old 
algorithm.  It generally won't change selections that are very short and 
it does well on slightly longer selections.  It is looking specifically 
for INCREASING slope AND close to zero.  Perhaps documentation should 
note that that is what it looks for.  If it mixed increasing and 
decreasing slope, you would get glitches.


> I thought we were planning to phase out selection at zero crossings - not least because it's pretty useless on stereo tracks.

Maybe.  But that is no reason to leave the bug open.

Sometime we SHOULD remove zero crossings, when we have something better, 
akin to beat-matching that works at a finer granularity, and that we can 
use on looped play and to repair joins and for stereo and so on.

--James.