Re: [Audacity-devel] NewTime Scale Effect : maximum/minimum tempo change

[Gale A. <ga...@au...>]

| From "Clayton Otey" <cla...@gm...> 
| Sat, 29 Nov 2008 13:41:42 -0800
| Subject: [Audacity-devel] NewTime Scale Effect : maximum/minimum tempo change
> Regarding the crash, did you also have a non-zero pitch shift?  I did a -91/-80 
> stretch and it didn't crash.  If you had a non-zero pitch shift, you may have 
> triggered an assert(stretch*pitch shift < 12, >1/12).  

No, I had zero pitch shift and it happened twice out of four times on
the same region of the same imported track.

> This shouldn't happen anyway, and I fixed the dialog so it doesn't allow the
> user to circumvent the parameter restrictions.

The new code does seem to prevent circumvention OK. Thanks. 
 
> I sort of arbitrarily decided on the parameter restrictions.  The processing time
> doesn't change very much with the parameters, so I didn't base my decision on 
> that.  It has more to do with quality.  I don't want to allow people to abuse the 
> algorithm and produce shameful results.  Feel free to change the restrictions, but 
> if you change them too much you'll have to change the sanity check asserts in
> lib-src/sbsms/src/sbsms.cpp as well.  These asserts are also arbitrary, so they can 
> be freely changed.

Based on what I've heard on reasonably typical pop and classical tracks,
I'd like to propose limits of -90% (-10x) and +500% (+6x). I think
that's a reasonable compromise of quality against not frustrating people
wanting to experiment. So if you agree, is all that needs to be done to 
change the values after "bool TimeScaleDialog::CheckParameters()" in
TimeScale.cpp, or do you think the asserts need changing? Note I had 
no crashing problems with -90/-80  or -90/-90 tempo changes after 
changing the limits to the above. 

I also compared a -10x constant change in Time Scale with one in  
Paul Nasca's "Paulstretch" extreme stretch effect (a patch for which
has been in our SF tracker for nearly a year):
http://sourceforge.net/tracker/index.php?func=detail&aid=1865955&group_id=6235&atid=306235

Paulstretch claims to be able to maintain high quality at stretches far
below -10x. I tried a standalone Windows executable of it. It did 
sound good at -10x on a solo female vocal with slowly changing string
and accordion chords and faster piano beats. Time Scale produced 
rather an unpleasant effect where the vocalist introduced more vibrato,
with the main and off-pitch notes stressed equally - Paulstretch treated
the off-pitch note more like a colouration with the main note always
prominent. On the other hand the piano beats were almost lost in
Paulstretch but very clear in Time Scale. Paulstretch has lots of 
configurations in its standalone version which I did not bother to 
explore. 


Gale