[Clam-devel] Spectral Analysis/Synthesis

[Karsten Krispin]

Hello!

Well, I'm new to this list, so let me first introduce a bit.. ;)

Currently I'm implementing an own synthesizer for our own reactable.
I'm using Qt4 for the graphical stuff while I chose CLAM for the DSP stuff.

The problem I have got, is that when I try to implement the spectral-domain 
stuff, I get quite disorted outputs when my signals has been pushed through a 
spectral analysis/synthesis-chain. 

My current setup in short is like the following:

 Samplerate: 48000
 Buffersize (Hopsize) : 512 ( I guess that's default)

Derived class from CLAM::Frame for Audio-Transport: I've just implemented some 
flags of whether the spectrum or audio array still holds the latest 
transformed data etc.. (Spectrum or Audio transforms are setting those flags 
automaticly, so that they know whether they need to synthesize the audio from 
spectrum or vice-versa).

Then, for my Synthesizer Items, I've derived QGraphicItems a dozent of times, 
but that should't matter. How ever, all Items who can perform such a 
conversation between audio and spectrum are having their own private 
SpectralAnalysis/SpectralSynthesis and OverlapAdd. That's due to the fact, 
that they run in different threads, for parallelism. And FFTW3 seems to need 
this, to become thread-safe. 

So, the init stuff for those three CLAM-Processing looks like the following 
(that's within the class-constructor):

	int samplingRate = 48000, frameSize = 512;
	
	CLAM::SpectralAnalysisConfig specAnalConf;
	CLAM::SpectralSynthesisConfig specSynthConf;
	CLAM::OverlapAddConfig	overlapConf;
	

	specAnalConf.SetSamplingRate(samplingRate);
	specAnalConf.SetHopSize(frameSize);
	this->_specAnal.Configure(specAnalConf);
	this->_specAnal.Start();
	

	specSynthConf.SetHopSize(frameSize);
	specSynthConf.SetSamplingRate(samplingRate);
	this->_specSynth.Configure(specSynthConf);
	this->_specSynth.Start();
	
	overlapConf.SetFrameSize(frameSize);
	this->_overlapAdd.Configure(overlapConf);
	this->_overlapAdd.Start();



Two functions of that class are responsible for the conversation:

AudioFlowItem::convertAudioToSpectrum(AudioPackage *package)
{
[...] this->_specAnal.Do(package->GetAudioFrame(), package->GetSpectrum())
}

AudioFlowItem::convertSpectrumToAudio(AudioPackage *package)
{

	[...]
	CLAM::Audio tmpAudio;
	tmpAudio.SetSize(package->frameSize()*2);
	this->_specSynth.Do(package->GetSpectrum(), tmpAudio);
	this->_overlapAdd.Do(tmpAudio, package->GetAudioFrame());
}


Whereas AudioPackage is the derived class of CLAM::Frame.
I've just included the relevant parts here, of course.


I'm sorry that I ask like a kind of "please code my application". But that 
what you see here is actually anything what fails within my app and I don't 
have any clue why... From what I understand, using the config-default's 
should work pretty well. I've also read the thread "Spectral 
Analysis/Synthesis problem" of globot on that list here. He seems to have/had 
the same problem. How ever, I could not get any hints out of that thread. 

To ramp up this mess, I've constructed the same part in NetworkEditor 
(SimpleOsc -> SpecAnal -> SpecSynth -> OverlapAdd -> AudioSink) and compared 
the config-values of Analysis and Synth  with those within my app. They are 
absolutely the same. But within NetworkEditor everything works fine.

If I leave out the spectrum stuff and just push out a pure sine-wave without 
anal/synth in between the generator and audio-sink, it will be outputted 
pretty smooth. So it is nothing driver/hw-related, like x-runs...

Thanks in advance and greetings from germany,

Karsten Krispin

Ps: Please excuse my bad english but it became late ;)