[Clam-devel] Pitch transposition network and a new network for pitch discretization

[Pau Arumi]

Hernan, all
Mistery solved :) and fix commited r10297

The offending commit was 10226. It was my fault, I applied Bernhard
Rettenbacher patch on SinTracking.cxx without enough insight:

$ svn diff -r10225:10226  CLAM/src/Processing/Analysis/SinTracking.cxx
Index: CLAM/src/Processing/Analysis/SinTracking.cxx
===================================================================
--- CLAM/src/Processing/Analysis/SinTracking.cxx        (revision 
10225)
+++ CLAM/src/Processing/Analysis/SinTracking.cxx        (revision 
10226)
@@ -226,6 +226,8 @@
    //xamat: test!
    SearchArray<TData> mySearch(peakFreqBuffer);
    TIndex found=mySearch.Find(currentPeakFreq);
+  if (found<0)
+         return found;
    TIndex originalFound = found;
    distance = Abs(peakFreqBuffer[found]-currentPeakFreq);
    //make sure that the two surrounding peaks are not in fact closer

This code belongs to the method
TIndex SinTracking::GetCandidate(TData currentPeakFreq, const 
SpectralPeakArray& iPeakArray, TData& distance) const
And the added early return prevented to execute necessary code even
in the case the first search found nothing.

In the same commit I also fixed the SMSTransposition.ui in-control
identifier. Probably it already was in Hernan's patch but not
applied correctly for some reason (I got some "rejects" while
patching)

Last, I discovered that you can not copy-paste in-control ids from
NE to designer because the bounds information are now also part of
the pasted text. I'm surprised nobody noticed this to the list.
I'll try to fix it soon --or better, apply a patch ;)

pau

En/na Hernán Ordiales ha escrit:
> On 6/22/07, Pau Arumi <parumi at iua.upf.edu> wrote:
> [snip]
>> The same here!
>> I've started a new network from scratch that uses SMSTranspose.
>> It seems to run fine when no control is sent. Once you start
>> sending control changes the audio doesn't change (transposition
>> does not work) and when control value gets around 1000 it crashes
>> CLAM::SynthSineSpectrum::FillSynthSineSpectrum
>>
>> The backtrace:
> [snip]
>> But there are no significant changes in SMSTranspose
>> from 1.1 release, only the factory registration changed.
> 
> could you told me more about the registration issue? then maybe i can
> figure it better, as you said seems to be the only change related to
> SMSSynthesis from version 1.1.0
> 
>> On the other side the sintracks plot out of smsanalysis already
>> shows this strange single-peak. The problem seems in smsanalysis,
>> however SMSAnalysisSynthesis back2back tests are green. I'm
>> puzzled...
>> Ideas, anyone?
>>
>> pau
> 
> I'm puzzled too
>