annotate dsp/tonal/ChangeDetectionFunction.cpp @ 73:dcb555b90924

* Key detector: when returning key strengths, use the peak value of the
  three underlying chromagram correlations (from 36-bin chromagram)
  corresponding to each key, instead of the mean.
  Rationale: This is the same method as used when returning the key value,
  and it's nice to have the same results in both returned value and plot.
  The peak performed better than the sum with a simple test set of triads,
  so it seems reasonable to change the plot to match the key output rather
  than the other way around.

* FFT: kiss_fftr returns only the non-conjugate bins, synthesise the rest
  rather than leaving them (perhaps dangerously) undefined.  Fixes an
  uninitialised data error in chromagram that could cause garbage results
  from key detector.

* Constant Q: remove precalculated values again, I reckon they're not
  proving such a good tradeoff.

/* -*- c-basic-offset: 4 indent-tabs-mode: nil -*-  vi:set ts=8 sts=4 sw=4: */

/*
    QM DSP Library

    Centre for Digital Music, Queen Mary, University of London.
    This file copyright 2006 Martin Gasser.
    All rights reserved.
*/

#include "ChangeDetectionFunction.h"

#ifndef PI
#define PI (3.14159265358979232846)
#endif



ChangeDetectionFunction::ChangeDetectionFunction(ChangeDFConfig config) :
	m_dFilterSigma(0.0), m_iFilterWidth(0)
{
	setFilterWidth(config.smoothingWidth);
}

ChangeDetectionFunction::~ChangeDetectionFunction()
{
}

void ChangeDetectionFunction::setFilterWidth(const int iWidth)
{
	m_iFilterWidth = iWidth*2+1;
	
	// it is assumed that the gaussian is 0 outside of +/- FWHM
	// => filter width = 2*FWHM = 2*2.3548*sigma
	m_dFilterSigma = double(m_iFilterWidth) / double(2*2.3548);
	m_vaGaussian.resize(m_iFilterWidth);
	
	double dScale = 1.0 / (m_dFilterSigma*sqrt(2*PI));
	
	for (int x = -(m_iFilterWidth-1)/2; x <= (m_iFilterWidth-1)/2; x++)
	{
		double w = dScale * std::exp ( -(x*x)/(2*m_dFilterSigma*m_dFilterSigma) );
		m_vaGaussian[x + (m_iFilterWidth-1)/2] = w;
	}
	
#ifdef DEBUG_CHANGE_DETECTION_FUNCTION
	std::cerr << "Filter sigma: " << m_dFilterSigma << std::endl;
	std::cerr << "Filter width: " << m_iFilterWidth << std::endl;
#endif
}


ChangeDistance ChangeDetectionFunction::process(const TCSGram& rTCSGram)
{
	ChangeDistance retVal;
	retVal.resize(rTCSGram.getSize(), 0.0);
	
	TCSGram smoothedTCSGram;

	for (int iPosition = 0; iPosition < rTCSGram.getSize(); iPosition++)
	{
		int iSkipLower = 0;
	
		int iLowerPos = iPosition - (m_iFilterWidth-1)/2;
		int iUpperPos = iPosition + (m_iFilterWidth-1)/2;
	
		if (iLowerPos < 0)
		{
			iSkipLower = -iLowerPos;
			iLowerPos = 0;
		}
	
		if (iUpperPos >= rTCSGram.getSize())
		{
			int iMaxIndex = rTCSGram.getSize() - 1;
			iUpperPos = iMaxIndex;
		}
	
		TCSVector smoothedVector;

		// for every bin of the vector, calculate the smoothed value
		for (int iPC = 0; iPC < 6; iPC++)
		{	
			size_t j = 0;
			double dSmoothedValue = 0.0;
			TCSVector rCV;
		
			for (int i = iLowerPos; i <= iUpperPos; i++)
			{
				rTCSGram.getTCSVector(i, rCV);
				dSmoothedValue += m_vaGaussian[iSkipLower + j++] * rCV[iPC];
			}

			smoothedVector[iPC] = dSmoothedValue;
		}
		
		smoothedTCSGram.addTCSVector(smoothedVector);
	}

	for (int iPosition = 0; iPosition < rTCSGram.getSize(); iPosition++)
	{
		/*
			TODO: calculate a confidence measure for the current estimation
			if the current estimate is not confident enough, look further into the future/the past
			e.g., High frequency content, zero crossing rate, spectral flatness
		*/
		
		TCSVector nextTCS;
		TCSVector previousTCS;
		
		int iWindow = 1;

		// while (previousTCS.magnitude() < 0.1 && (iPosition-iWindow) > 0)
		{
			smoothedTCSGram.getTCSVector(iPosition-iWindow, previousTCS);
			// std::cout << previousTCS.magnitude() << std::endl;
			iWindow++;
		}
		
		iWindow = 1;
		
		// while (nextTCS.magnitude() < 0.1 && (iPosition+iWindow) < (rTCSGram.getSize()-1) )
		{
			smoothedTCSGram.getTCSVector(iPosition+iWindow, nextTCS);
			iWindow++;
		}

		double distance = 0.0;
		// Euclidean distance
		for (size_t j = 0; j < 6; j++)
		{
			distance += std::pow(nextTCS[j] - previousTCS[j], 2.0);
		}
	
		retVal[iPosition] = std::pow(distance, 0.5);
	}

	return retVal;
}