[Clam-devel] [PATCH] CardinalVowel Control, doxygen documentation for formant extractor

[abe kazemzadeh]

Hi Pau and David,

thanks for the instructions on how to get the processings registered.  I'll
do that in the future.  Also, thanks for taking the screen shot.  Maybe in
future iterations I can add some flashy widgets :)

> Do you think it will be easy to improve the vowel model? Does not
> sound very human, right?

Yep, there are several things that I think might help.  I think that the
harmonic peak generator could use the most improvement.  I played around
with the controls on the simple oscillator that feeds into the residual
spectrum of the SMSSynthesis (SimpleOscillator->spectralAnalysis->residual
spectrum port of SMSSynthesis).  Using low values for pitch and amplitude
seems to add some noise between the harmonics and makes it sound a little
better.  That's just fiddling around, but I think that the purely harmonic
peaks are the reason it's unnatural sounding, so I'll look into this
and see if I can find a description of the spectrum of the vocal fold
source.
Also I think adding jitter and shimmer (slight variations in the pitch and
amplitude) would help it sound more natural.  Right now I'm working on an
ADSR envelope to output syllable-like pulses so that might help, but I think
that's not the main inhuman aspect.  Another thing is to
take formant bandwidth into account in the vowel resonator.  I think that
this is also not the main cause of the unnatural sound, but I could be
wrong, so I'll check out this if the other things don't work.

If anyone has any ideas about how to best do any of these improvements, let
me know!

Also, I wanted to ask about the ADSR processing.  It seems that the ladspa
ADSR processing is has more features than clam's (gates/triggers).  I'm not
sure if it would be good to add the extra features to clam's ADSR processing
or just use the  laspa one?  Also it could be that I"m not understanding how
the ADSR processing is supposed to be used in clam.

Thanks,
Abe

PS, I'm currently working on clam as I'm taking a cross country trip.  Not
having the internet (and the plains states) have been good
for productivity, but I can only respond to emails at night when I'm
at a wifi hotel (it's amazing how many small hotels in little towns
have wireless internet).  I guess it shouldn't be much of a problem since
we're all in our own time zones, but just in case you wonder why I don't
reply during the daytime hours...


On 8/16/07, David García Garzón <dgarcia at iua.upf.edu> wrote:
>
> On Thursday 16 August 2007 19:14:31 Pau Arumi wrote:
> > En/na Pau Arumi ha escrit:
> > > Notice the cathegory "Speech (Experimental - GSoC)" in the
> > > processing tree. These processings have been commited in
> > > CLAM/plugins/speech. Take a look how (easily) they are registed
> > > with metadata. You can expand the description, if you wish.
> >
> > A quick reminder:
> > export CLAM_PLUGIN_PATH=/path/to/plugins/speech/
> > before executing NE
>
> As a more general solution plugins SConstruct install plugins in
> clam_install_prefix/lib/clam whenever you do 'scons install'. If you setup
> CLAM_PLUGIN_PATH to such path in your environment you can forget about
> making
> it pointing different places. By (locally) installing the plugin you get
> it.
>
> David.
>
>
> _______________________________________________
> Clam-devel mailing list
> Clam-devel at llistes.projectes.lafarga.org
> https://llistes.projectes.lafarga.org/cgi-bin/mailman/listinfo/clam-devel
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.clam-project.org/pipermail/clam-devel-clam-project.org/attachments/20070817/0cdac9b4/attachment-0003.htm>