cannam@0: /* -*- c-basic-offset: 4 indent-tabs-mode: nil -*-  vi:set ts=8 sts=4 sw=4: */
cannam@0: 
cannam@0: /*
cannam@0:     QM DSP Library
cannam@0: 
cannam@0:     Centre for Digital Music, Queen Mary, University of London.
cannam@0:     This file copyright 2006 Martin Gasser.
Chris@84: 
Chris@84:     This program is free software; you can redistribute it and/or
Chris@84:     modify it under the terms of the GNU General Public License as
Chris@84:     published by the Free Software Foundation; either version 2 of the
Chris@84:     License, or (at your option) any later version.  See the file
Chris@84:     COPYING included with this distribution for more information.
cannam@0: */
cannam@0: 
cannam@0: #include "ChangeDetectionFunction.h"
cannam@0: 
cannam@0: #ifndef PI
cannam@0: #define PI (3.14159265358979232846)
cannam@0: #endif
cannam@0: 
cannam@0: 
cannam@0: 
cannam@0: ChangeDetectionFunction::ChangeDetectionFunction(ChangeDFConfig config) :
cannam@0: 	m_dFilterSigma(0.0), m_iFilterWidth(0)
cannam@0: {
cannam@0: 	setFilterWidth(config.smoothingWidth);
cannam@0: }
cannam@0: 
cannam@0: ChangeDetectionFunction::~ChangeDetectionFunction()
cannam@0: {
cannam@0: }
cannam@0: 
cannam@0: void ChangeDetectionFunction::setFilterWidth(const int iWidth)
cannam@0: {
cannam@0: 	m_iFilterWidth = iWidth*2+1;
cannam@0: 	
cannam@0: 	// it is assumed that the gaussian is 0 outside of +/- FWHM
cannam@0: 	// => filter width = 2*FWHM = 2*2.3548*sigma
cannam@0: 	m_dFilterSigma = double(m_iFilterWidth) / double(2*2.3548);
cannam@0: 	m_vaGaussian.resize(m_iFilterWidth);
cannam@0: 	
cannam@0: 	double dScale = 1.0 / (m_dFilterSigma*sqrt(2*PI));
cannam@0: 	
cannam@0: 	for (int x = -(m_iFilterWidth-1)/2; x <= (m_iFilterWidth-1)/2; x++)
cannam@0: 	{
cannam@0: 		double w = dScale * std::exp ( -(x*x)/(2*m_dFilterSigma*m_dFilterSigma) );
cannam@0: 		m_vaGaussian[x + (m_iFilterWidth-1)/2] = w;
cannam@0: 	}
cannam@0: 	
cannam@0: #ifdef DEBUG_CHANGE_DETECTION_FUNCTION
cannam@50: 	std::cerr << "Filter sigma: " << m_dFilterSigma << std::endl;
cannam@50: 	std::cerr << "Filter width: " << m_iFilterWidth << std::endl;
cannam@0: #endif
cannam@0: }
cannam@0: 
cannam@0: 
cannam@0: ChangeDistance ChangeDetectionFunction::process(const TCSGram& rTCSGram)
cannam@0: {
cannam@0: 	ChangeDistance retVal;
cannam@0: 	retVal.resize(rTCSGram.getSize(), 0.0);
cannam@0: 	
cannam@0: 	TCSGram smoothedTCSGram;
cannam@0: 
cannam@0: 	for (int iPosition = 0; iPosition < rTCSGram.getSize(); iPosition++)
cannam@0: 	{
cannam@0: 		int iSkipLower = 0;
cannam@0: 	
cannam@0: 		int iLowerPos = iPosition - (m_iFilterWidth-1)/2;
cannam@0: 		int iUpperPos = iPosition + (m_iFilterWidth-1)/2;
cannam@0: 	
cannam@0: 		if (iLowerPos < 0)
cannam@0: 		{
cannam@0: 			iSkipLower = -iLowerPos;
cannam@0: 			iLowerPos = 0;
cannam@0: 		}
cannam@0: 	
cannam@0: 		if (iUpperPos >= rTCSGram.getSize())
cannam@0: 		{
cannam@0: 			int iMaxIndex = rTCSGram.getSize() - 1;
cannam@0: 			iUpperPos = iMaxIndex;
cannam@0: 		}
cannam@0: 	
cannam@0: 		TCSVector smoothedVector;
cannam@0: 
cannam@0: 		// for every bin of the vector, calculate the smoothed value
cannam@0: 		for (int iPC = 0; iPC < 6; iPC++)
cannam@0: 		{	
cannam@0: 			size_t j = 0;
cannam@0: 			double dSmoothedValue = 0.0;
cannam@0: 			TCSVector rCV;
cannam@0: 		
cannam@0: 			for (int i = iLowerPos; i <= iUpperPos; i++)
cannam@0: 			{
cannam@0: 				rTCSGram.getTCSVector(i, rCV);
cannam@0: 				dSmoothedValue += m_vaGaussian[iSkipLower + j++] * rCV[iPC];
cannam@0: 			}
cannam@0: 
cannam@0: 			smoothedVector[iPC] = dSmoothedValue;
cannam@0: 		}
cannam@0: 		
cannam@0: 		smoothedTCSGram.addTCSVector(smoothedVector);
cannam@0: 	}
cannam@0: 
cannam@0: 	for (int iPosition = 0; iPosition < rTCSGram.getSize(); iPosition++)
cannam@0: 	{
cannam@0: 		/*
cannam@0: 			TODO: calculate a confidence measure for the current estimation
cannam@0: 			if the current estimate is not confident enough, look further into the future/the past
cannam@0: 			e.g., High frequency content, zero crossing rate, spectral flatness
cannam@0: 		*/
cannam@0: 		
cannam@0: 		TCSVector nextTCS;
cannam@0: 		TCSVector previousTCS;
cannam@0: 		
cannam@0: 		int iWindow = 1;
cannam@0: 
cannam@0: 		// while (previousTCS.magnitude() < 0.1 && (iPosition-iWindow) > 0)
cannam@0: 		{
cannam@0: 			smoothedTCSGram.getTCSVector(iPosition-iWindow, previousTCS);
cannam@0: 			// std::cout << previousTCS.magnitude() << std::endl;
cannam@0: 			iWindow++;
cannam@0: 		}
cannam@0: 		
cannam@0: 		iWindow = 1;
cannam@0: 		
cannam@0: 		// while (nextTCS.magnitude() < 0.1 && (iPosition+iWindow) < (rTCSGram.getSize()-1) )
cannam@0: 		{
cannam@0: 			smoothedTCSGram.getTCSVector(iPosition+iWindow, nextTCS);
cannam@0: 			iWindow++;
cannam@0: 		}
cannam@0: 
cannam@0: 		double distance = 0.0;
cannam@0: 		// Euclidean distance
cannam@0: 		for (size_t j = 0; j < 6; j++)
cannam@0: 		{
cannam@0: 			distance += std::pow(nextTCS[j] - previousTCS[j], 2.0);
cannam@0: 		}
cannam@0: 	
cannam@0: 		retVal[iPosition] = std::pow(distance, 0.5);
cannam@0: 	}
cannam@0: 
cannam@0: 	return retVal;
cannam@0: }