[Clam-devel] streaming SDIF files

Greg Kellum greg.kellum at gmail.com
Fri Jun 29 06:43:49 PDT 2007


Hi Xavier and Pau,

There actually shouldn't be any problem with the configuration of
SMSSynthesis.  I dumped the configuration of the SMSAnalysisCore and
SMSSynthesis objects in the SMSTools application to XML files, and I have
used these files to resynthesize a WAV from an SDIF file without any
problems when I was using Segments.  The problem I mentioned just sprung up
when I started streaming the SDIF file.  But nonetheless, I just tried a
couple of alternative hop sizes, but it didn't help.  When you look at the
resynthesized output in an audio editor, one sees sudden jumps /
discontinuities in the audio signal which would imply that a triangular
window isn't being applied to the signal, because if it was, the signal
would be 0 at the window boundaries.  I'm looking at SMSSynthesis though at
the moment to get a better idea of what is happening...

...

I just got Pau's mail.  The problem you had running the example happened
because it couldn't find the configuration file synthesis.xml.  Just move
that file out of the directory SDIFToWavStreaming into its parent directory,
and it will find it, or give the relative path to the file:
SDIFToWavStreaming/synthesis.xml.  I did that myself locally, but I forgot
to add that moved synthesis.xml file to the SVN repository.

By the way so people in CLAM use tabs, not spaces?  I thought it was the
other way around...

Best,
Greg



On 6/29/07, Xavier Amatriain <xavier at create.ucsb.edu> wrote:
>
> Hi Greg,
>
> > I've been having a problem with artifacts though that seem to be
> > coming from the overlapping of windows.  My guess is that either a
> > triangular window is not being applied to the audio windows before
> > they are being overlapped or that the step-size is not correct.
> I am almost sure that your problem is with the hop/size. If you use
> SMSSynthesis it is very unlikely that the triangular window is not
> applied.
> > But even though I've played around with a lot of different parameter
> > values, I can't seem to make the problem go away.  Can someone tell me
> > where exactly this overlap-adding is supposed to be taking place?
> > Which class is responsible for it?
> The OverlapAdd takes place in the SMSSynthesis class lines 260 and 268:
>
>
> http://www.clam.iua.upf.edu/doc/CLAM-devel-doxygen/SMSSynthesis_8cxx-source.html#l00200
> ;
>
> And the class responsible for it is the OverlapAdd class
>
>
> http://www.clam.iua.upf.edu/doc/CLAM-devel-doxygen/classCLAM_1_1OverlapAdd.html
>
> A rule of thumb: synthesis hopsize and framesize should both be equal to
> analysis hopsize. (Synthesis hopsize defines how much the
> triangular window advances and framesize is half of the triangular
> window size. Note that neither of them have anything to do with the
> analysis window size).
>
> Hope it helps.
>
> _______________________________________________
> 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/20070629/cd11c5a5/attachment-0004.htm>


More information about the clam-devel mailing list