Mercurial > hg > vamp-tempogram

--- a/NoveltyCurveProcessor.cpp	Mon Aug 18 15:22:44 2014 +0100
+++ b/NoveltyCurveProcessor.cpp	Tue Aug 19 16:52:19 2014 +0100
@@ -11,11 +11,10 @@
 #include "NoveltyCurveProcessor.h"
 using namespace std;

-NoveltyCurveProcessor::NoveltyCurveProcessor(const float &samplingFrequency, const size_t &fftLength, const size_t &numberOfBlocks, const size_t &compressionConstant) :
+NoveltyCurveProcessor::NoveltyCurveProcessor(const float &samplingFrequency, const size_t &fftLength,  const size_t &compressionConstant) :
     m_samplingFrequency(samplingFrequency),
     m_fftLength(fftLength),
     m_blockSize(fftLength/2 + 1),
-    m_numberOfBlocks(numberOfBlocks),
     m_compressionConstant(compressionConstant),
     m_numberOfBands(5),
     m_pBandBoundaries(0),
@@ -54,12 +53,15 @@
 }

 //calculate max of spectrogram
-float NoveltyCurveProcessor::calculateMax(const vector< vector<float> > &spectrogram) const
+float NoveltyCurveProcessor::calculateMax(const Spectrogram &spectrogram) const
 {
     float max = 0;

-    for (unsigned int j = 0; j < m_numberOfBlocks; j++){
-        for (unsigned int i = 0; i < m_blockSize; i++){
+    int length = spectrogram.size();
+    int height = spectrogram[0].size();
+
+    for (int i = 0; i < length; i++){
+        for (int j = 0; j < height; j++){
             max = max > fabs(spectrogram[i][j]) ? max : fabs(spectrogram[i][j]);
         }
     }
@@ -71,16 +73,16 @@
 //uses m_hannWindow as filter
 void NoveltyCurveProcessor::subtractLocalAverage(vector<float> &noveltyCurve, const size_t &smoothLength) const
 {
-    vector<float> localAverage(m_numberOfBlocks);
+    int numberOfBlocks = noveltyCurve.size();
+    vector<float> localAverage(numberOfBlocks);

     float * m_hannWindow = new float[smoothLength];
     WindowFunction::hanning(m_hannWindow, smoothLength, true);

-    FIRFilter filter(m_numberOfBlocks, smoothLength);
+    FIRFilter filter(numberOfBlocks, smoothLength);
     filter.process(&noveltyCurve[0], m_hannWindow, &localAverage[0], FIRFilter::middle);

-    assert(noveltyCurve.size() == m_numberOfBlocks);
-    for (unsigned int i = 0; i < m_numberOfBlocks; i++){
+    for (int i = 0; i < numberOfBlocks; i++){
         noveltyCurve[i] -= localAverage[i];
         noveltyCurve[i] = noveltyCurve[i] >= 0 ? noveltyCurve[i] : 0;
     }
@@ -90,9 +92,11 @@
 }

 //smoothed differentiator filter. Flips upper half of hanning window about y-axis to create coefficients.
-void NoveltyCurveProcessor::smoothedDifferentiator(vector< vector<float> > &spectrogram, const size_t &smoothLength) const
+void NoveltyCurveProcessor::smoothedDifferentiator(SpectrogramTransposed &spectrogramTransposed, const size_t &smoothLength) const
 {

+    int numberOfBlocks = spectrogramTransposed[0].size();
+
     float * diffHannWindow = new float [smoothLength];
     WindowFunction::hanning(diffHannWindow, smoothLength, true);

@@ -101,56 +105,56 @@
         diffHannWindow[i] = -diffHannWindow[i];
     }

-    FIRFilter smoothFilter(m_numberOfBlocks, smoothLength);
+    FIRFilter smoothFilter(numberOfBlocks, smoothLength);

     for (int i = 0; i < (int)m_blockSize; i++){
-        smoothFilter.process(&spectrogram[i][0], diffHannWindow, &spectrogram[i][0], FIRFilter::middle);
+        smoothFilter.process(&spectrogramTransposed[i][0], diffHannWindow, &spectrogramTransposed[i][0], FIRFilter::middle);
     }
 }

 //half rectification (set negative to zero)
-void NoveltyCurveProcessor::halfWaveRectify(vector< vector<float> > &spectrogram) const
+void NoveltyCurveProcessor::halfWaveRectify(SpectrogramTransposed &spectrogramTransposed) const
 {
-    for (int block = 0; block < (int)m_numberOfBlocks; block++){
+    int numberOfBlocks = spectrogramTransposed[0].size();
+
+    for (int block = 0; block < (int)numberOfBlocks; block++){
         for (int k = 0; k < (int)m_blockSize; k++){
-            if (spectrogram[k][block] < 0.0) spectrogram[k][block] = 0.0;
+            if (spectrogramTransposed[k][block] < 0.0) spectrogramTransposed[k][block] = 0.0;
         }
     }
 }

 //process method
 vector<float>
-NoveltyCurveProcessor::spectrogramToNoveltyCurve(Spectrogram spectrogram) const
+NoveltyCurveProcessor::spectrogramToNoveltyCurve(const Spectrogram &spectrogram) const //make argument const &
 {
-    std::vector<float> noveltyCurve(m_numberOfBlocks);
-
-    //cout << spectrogram[0].size() << " : " << m_numberOfBlocks << endl;
-    assert(spectrogram.size() == m_blockSize);
-    assert(spectrogram[0].size() == m_numberOfBlocks);
+    int numberOfBlocks = spectrogram.size();
+    std::vector<float> noveltyCurve(numberOfBlocks);
+    SpectrogramTransposed spectrogramTransposed(spectrogram[0].size(), vector<float>(spectrogram.size()));

     //normalise and log spectrogram
     float normaliseScale = calculateMax(spectrogram);
-    for (int block = 0; block < (int)m_numberOfBlocks; block++){
+    for (int block = 0; block < (int)numberOfBlocks; block++){
         for (int k = 0; k < (int)m_blockSize; k++){
-            if(normaliseScale != 0.0) spectrogram[k][block] /= normaliseScale; //normalise
-            spectrogram[k][block] = log(1+m_compressionConstant*spectrogram[k][block]);
+            spectrogramTransposed[k][block] = log(1+m_compressionConstant*spectrogram[block][k]);
+            if(normaliseScale != 0.0) spectrogramTransposed[k][block] /= normaliseScale; //normalise
         }
     }

     //smooted differentiator
-    smoothedDifferentiator(spectrogram, 5); //make smoothLength a parameter!
+    smoothedDifferentiator(spectrogramTransposed, 5); //make smoothLength a parameter!
     //halfwave rectification
-    halfWaveRectify(spectrogram);
+    halfWaveRectify(spectrogramTransposed);

     //bandwise processing
-    for (int block = 0; block < (int)m_numberOfBlocks; block++){
+    for (int block = 0; block < (int)numberOfBlocks; block++){
         for (int band = 0; band < (int)m_numberOfBands; band++){
             int k = m_pBandBoundaries[band];
             int bandEnd = m_pBandBoundaries[band+1];
             m_pBandSum[band] = 0;

             while(k < bandEnd){
-                m_pBandSum[band] += spectrogram[k][block];
+                m_pBandSum[band] += spectrogramTransposed[k][block];
                 k++;
             }
         }
--- a/NoveltyCurveProcessor.h	Mon Aug 18 15:22:44 2014 +0100
+++ b/NoveltyCurveProcessor.h	Tue Aug 19 16:52:19 2014 +0100
@@ -24,7 +24,6 @@
     float m_samplingFrequency;
     size_t m_fftLength;
     size_t m_blockSize;
-    size_t m_numberOfBlocks;
     int m_compressionConstant;
     size_t m_numberOfBands;
     int * m_pBandBoundaries;
@@ -32,16 +31,16 @@

     void initialise();
     void cleanup();
-    float calculateMax(const std::vector< std::vector<float> > &spectrogram) const;
+    float calculateMax(const Spectrogram &spectrogram) const;
     void subtractLocalAverage(std::vector<float> &noveltyCurve, const size_t &smoothLength) const;
-    void smoothedDifferentiator(std::vector< std::vector<float> > &spectrogram, const size_t &smoothLength) const;
-    void halfWaveRectify(std::vector< std::vector<float> > &spectrogram) const;
+    void smoothedDifferentiator(SpectrogramTransposed &spectrogram, const size_t &smoothLength) const;
+    void halfWaveRectify(SpectrogramTransposed &spectrogram) const;

 public:

-    NoveltyCurveProcessor(const float &samplingFrequency, const size_t &fftLength, const size_t &numberOfBlocks, const size_t &compressionConstant);
+    NoveltyCurveProcessor(const float &samplingFrequency, const size_t &fftLength, const size_t &compressionConstant);
     ~NoveltyCurveProcessor();
-    std::vector<float> spectrogramToNoveltyCurve(Spectrogram spectrogram) const;
+    std::vector<float> spectrogramToNoveltyCurve(const Spectrogram &spectrogram) const;
 };

 #endif /* defined(__Tempogram__NoveltyCurve__) */
--- a/TempogramPlugin.cpp	Mon Aug 18 15:22:44 2014 +0100
+++ b/TempogramPlugin.cpp	Tue Aug 19 16:52:19 2014 +0100
@@ -312,14 +312,6 @@
 {
 }

-string TempogramPlugin::floatToString(float value) const
-{
-    ostringstream ss;
-
-    if(!(ss << value)) throw runtime_error("TempogramPlugin::floatToString(): invalid conversion from float to string");
-    return ss.str();
-}
-
 TempogramPlugin::OutputList
 TempogramPlugin::getOutputDescriptors() const
 {
@@ -383,34 +375,6 @@
     return list;
 }

-bool TempogramPlugin::handleParameterValues(){
-
-    if (m_tempogramHopSize <= 0) return false;
-    if (m_tempogramLog2FftLength <= 0) return false;
-
-    if (m_tempogramFftLength < m_tempogramWindowLength){
-        m_tempogramFftLength = m_tempogramWindowLength;
-    }
-    if (m_tempogramMinBPM > m_tempogramMaxBPM){
-        m_tempogramMinBPM = 30;
-        m_tempogramMaxBPM = 480;
-    }
-
-    float tempogramInputSampleRate = (float)m_inputSampleRate/m_inputStepSize;
-    m_tempogramMinBin = (max(floor(((m_tempogramMinBPM/60)/tempogramInputSampleRate)*m_tempogramFftLength), (float)0.0));
-    m_tempogramMaxBin = (min(ceil(((m_tempogramMaxBPM/60)/tempogramInputSampleRate)*m_tempogramFftLength), (float)m_tempogramFftLength/2));
-
-    if (m_tempogramMinBPM > m_cyclicTempogramMinBPM) m_cyclicTempogramMinBPM = m_tempogramMinBPM;
-    float cyclicTempogramMaxBPM = 480;
-    if (m_tempogramMaxBPM < cyclicTempogramMaxBPM) cyclicTempogramMaxBPM = m_tempogramMaxBPM;
-
-    m_cyclicTempogramNumberOfOctaves = floor(log2(cyclicTempogramMaxBPM/m_cyclicTempogramMinBPM));
-    int numberOfBinsInFirstOctave = bpmToBin(m_cyclicTempogramMinBPM);
-    if (m_cyclicTempogramOctaveDivider > numberOfBinsInFirstOctave) m_cyclicTempogramOctaveDivider = numberOfBinsInFirstOctave;
-
-    return true;
-}
-
 bool
 TempogramPlugin::initialise(size_t channels, size_t stepSize, size_t blockSize)
 {
@@ -421,7 +385,6 @@
     m_inputBlockSize = blockSize;
     m_inputStepSize = stepSize;

-    m_spectrogram = SpectrogramTransposed(m_inputBlockSize/2.0f + 1);
     if (!handleParameterValues()) return false;
     //cout << m_cyclicTempogramOctaveDivider << endl;

@@ -433,7 +396,6 @@
 {
     // Clear buffers, reset stored values, etc
     m_spectrogram.clear();
-    m_spectrogram = SpectrogramTransposed(m_inputBlockSize/2.0f + 1);
     handleParameterValues();
 }

@@ -450,51 +412,17 @@
     const float *in = inputBuffers[0];

     //calculate magnitude of FrequencyDomain input
+    vector<float> fftCoefficients;
     for (int i = 0; i < (int)n; i++){
         float magnitude = sqrt(in[2*i] * in[2*i] + in[2*i + 1] * in[2*i + 1]);
         magnitude = magnitude > m_noveltyCurveMinDB ? magnitude : m_noveltyCurveMinDB;
-        m_spectrogram[i].push_back(magnitude);
+        fftCoefficients.push_back(magnitude);
     }
+    m_spectrogram.push_back(fftCoefficients);

     return featureSet;
 }

-vector<unsigned int> TempogramPlugin::calculateTempogramNearestNeighbourLogBins() const
-{
-    vector<unsigned int> logBins;
-
-    for (int i = 0; i < (int)ceil(m_cyclicTempogramNumberOfOctaves*m_cyclicTempogramOctaveDivider); i++){
-        float bpm = m_cyclicTempogramMinBPM*pow(2.0f, (float)i/m_cyclicTempogramOctaveDivider);
-        int bin = bpmToBin(bpm);
-
-        logBins.push_back(bin);
-        //cerr << bin << endl;
-    }
-
-    //cerr << logBins.size() << endl;
-
-    return logBins;
-}
-
-int TempogramPlugin::bpmToBin(const float &bpm) const
-{
-    float w = (float)bpm/60;
-    float sampleRate = m_inputSampleRate/m_inputStepSize;
-    int bin = floor((float)m_tempogramFftLength*w/sampleRate + 0.5);
-
-    if(bin < 0) bin = 0;
-    else if(bin > m_tempogramFftLength/2.0f) bin = m_tempogramFftLength;
-
-    return bin;
-}
-
-float TempogramPlugin::binToBPM(const int &bin) const
-{
-    float sampleRate = m_inputSampleRate/m_inputStepSize;
-
-    return (bin*sampleRate/m_tempogramFftLength)*60;
-}
-
 TempogramPlugin::FeatureSet
 TempogramPlugin::getRemainingFeatures()
 {
@@ -507,9 +435,9 @@
     FeatureSet featureSet;

     //initialise novelty curve processor
-    size_t numberOfBlocks = m_spectrogram[0].size();
+    size_t numberOfBlocks = m_spectrogram.size();
     //cerr << numberOfBlocks << endl;
-    NoveltyCurveProcessor nc(m_inputSampleRate, m_inputBlockSize, numberOfBlocks, m_noveltyCurveCompressionConstant);
+    NoveltyCurveProcessor nc(m_inputSampleRate, m_inputBlockSize, m_noveltyCurveCompressionConstant);
     vector<float> noveltyCurve = nc.spectrogramToNoveltyCurve(m_spectrogram); //calculate novelty curvefrom magnitude data
     //if(noveltyCurve.size() > 50) for (int i = 0; i < 50; i++) cerr << noveltyCurve[i] << endl;

@@ -548,20 +476,17 @@
     }

     //Calculate cyclic tempogram
-    vector<unsigned int> logBins = calculateTempogramNearestNeighbourLogBins();
+    vector< vector<unsigned int> > logBins = calculateTempogramNearestNeighbourLogBins();

-    assert(logBins.back() <= m_tempogramFftLength/2.0f);
-    assert((int)logBins.size() == m_cyclicTempogramOctaveDivider*m_cyclicTempogramNumberOfOctaves);
+    //assert((int)logBins.size() == m_cyclicTempogramOctaveDivider*m_cyclicTempogramNumberOfOctaves);
     for (int block = 0; block < tempogramLength; block++){
         Feature cyclicTempogramFeature;

         for (int i = 0; i < (int)m_cyclicTempogramOctaveDivider; i++){
             float sum = 0;

-            //mcerr << floor(binToBPM(logBins[i]) + 0.5) << " " << floor(binToBPM(logBins[i + m_cyclicTempogramOctaveDivider]) + 0.5) << endl;
-
             for (int j = 0; j < (int)m_cyclicTempogramNumberOfOctaves; j++){
-                sum += tempogram[block][logBins[i+j*m_cyclicTempogramOctaveDivider]];
+                sum += tempogram[block][logBins[j][i]];
             }
             cyclicTempogramFeature.values.push_back(sum/m_cyclicTempogramNumberOfOctaves);
             assert(!isnan(cyclicTempogramFeature.values.back()));
@@ -573,3 +498,76 @@

     return featureSet;
 }
+
+vector< vector<unsigned int> > TempogramPlugin::calculateTempogramNearestNeighbourLogBins() const
+{
+    vector< vector<unsigned int> > logBins;
+
+    for (int octave = 0; octave < (int)m_cyclicTempogramNumberOfOctaves; octave++){
+        vector<unsigned int> octaveBins;
+
+        for (int bin = 0; bin < (int)m_cyclicTempogramOctaveDivider; bin++){
+            float bpm = m_cyclicTempogramMinBPM*pow(2.0f, octave+(float)bin/m_cyclicTempogramOctaveDivider);
+
+            octaveBins.push_back(bpmToBin(bpm));
+        }
+        logBins.push_back(octaveBins);
+    }
+
+    //cerr << logBins.size() << endl;
+
+    return logBins;
+}
+
+unsigned int TempogramPlugin::bpmToBin(const float &bpm) const
+{
+    float w = (float)bpm/60;
+    float sampleRate = m_inputSampleRate/m_inputStepSize;
+    int bin = floor((float)m_tempogramFftLength*w/sampleRate + 0.5);
+
+    if(bin < 0) bin = 0;
+    else if(bin > m_tempogramFftLength/2.0f) bin = m_tempogramFftLength;
+
+    return bin;
+}
+
+float TempogramPlugin::binToBPM(const int &bin) const
+{
+    float sampleRate = m_inputSampleRate/m_inputStepSize;
+
+    return (bin*sampleRate/m_tempogramFftLength)*60;
+}
+
+bool TempogramPlugin::handleParameterValues(){
+
+    if (m_tempogramHopSize <= 0) return false;
+    if (m_tempogramLog2FftLength <= 0) return false;
+
+    if (m_tempogramFftLength < m_tempogramWindowLength){
+        m_tempogramFftLength = m_tempogramWindowLength;
+    }
+    if (m_tempogramMinBPM >= m_tempogramMaxBPM){
+        m_tempogramMinBPM = 30;
+        m_tempogramMaxBPM = 480;
+    }
+
+    float tempogramInputSampleRate = (float)m_inputSampleRate/m_inputStepSize;
+    m_tempogramMinBin = (max(floor(((m_tempogramMinBPM/60)/tempogramInputSampleRate)*m_tempogramFftLength), (float)0.0));
+    m_tempogramMaxBin = (min(ceil(((m_tempogramMaxBPM/60)/tempogramInputSampleRate)*m_tempogramFftLength), (float)m_tempogramFftLength/2));
+
+    if (m_tempogramMinBPM > m_cyclicTempogramMinBPM) m_cyclicTempogramMinBPM = m_tempogramMinBPM;
+    float cyclicTempogramMaxBPM = 480;
+    if (m_tempogramMaxBPM < cyclicTempogramMaxBPM) cyclicTempogramMaxBPM = m_tempogramMaxBPM;
+
+    m_cyclicTempogramNumberOfOctaves = floor(log2(cyclicTempogramMaxBPM/m_cyclicTempogramMinBPM));
+
+    return true;
+}
+
+string TempogramPlugin::floatToString(float value) const
+{
+    ostringstream ss;
+
+    if(!(ss << value)) throw runtime_error("TempogramPlugin::floatToString(): invalid conversion from float to string");
+    return ss.str();
+}
--- a/TempogramPlugin.h	Mon Aug 18 15:22:44 2014 +0100
+++ b/TempogramPlugin.h	Tue Aug 19 16:52:19 2014 +0100
@@ -77,7 +77,7 @@
     // plugin-specific data and methods go here
     size_t m_inputBlockSize;
     size_t m_inputStepSize;
-    SpectrogramTransposed m_spectrogram; //spectrogram data
+    Spectrogram m_spectrogram; //spectrogram data

     //Novelty Curve specific parameters
     float m_noveltyCurveMinDB;
@@ -102,8 +102,8 @@
     int m_cyclicTempogramOctaveDivider;

     string floatToString(float value) const;
-    vector<unsigned int> calculateTempogramNearestNeighbourLogBins() const;
-    int bpmToBin(const float &bpm) const;
+    vector< vector<unsigned int> > calculateTempogramNearestNeighbourLogBins() const;
+    unsigned int bpmToBin(const float &bpm) const;
     float binToBPM (const int &bin) const;
     bool handleParameterValues();
 };