cannam@198: /* -*- c-basic-offset: 4 indent-tabs-mode: nil -*-  vi:set ts=8 sts=4 sw=4: */
cannam@198: 
cannam@198: /*
cannam@198:     Vamp
cannam@198: 
cannam@198:     An API for audio analysis and feature extraction plugins.
cannam@198: 
cannam@198:     Centre for Digital Music, Queen Mary, University of London.
cannam@198:     Copyright 2006-2008 Chris Cannam and QMUL.
cannam@198:   
cannam@198:     Permission is hereby granted, free of charge, to any person
cannam@198:     obtaining a copy of this software and associated documentation
cannam@198:     files (the "Software"), to deal in the Software without
cannam@198:     restriction, including without limitation the rights to use, copy,
cannam@198:     modify, merge, publish, distribute, sublicense, and/or sell copies
cannam@198:     of the Software, and to permit persons to whom the Software is
cannam@198:     furnished to do so, subject to the following conditions:
cannam@198: 
cannam@198:     The above copyright notice and this permission notice shall be
cannam@198:     included in all copies or substantial portions of the Software.
cannam@198: 
cannam@198:     THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
cannam@198:     EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
cannam@198:     MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
cannam@198:     NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR
cannam@198:     ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF
cannam@198:     CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
cannam@198:     WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
cannam@198: 
cannam@198:     Except as contained in this notice, the names of the Centre for
cannam@198:     Digital Music; Queen Mary, University of London; and Chris Cannam
cannam@198:     shall not be used in advertising or otherwise to promote the sale,
cannam@198:     use or other dealings in this Software without prior written
cannam@198:     authorization.
cannam@198: */
cannam@198: 
cannam@198: #include "FixedTempoEstimator.h"
cannam@198: 
cannam@198: using std::string;
cannam@198: using std::vector;
cannam@198: using std::cerr;
cannam@198: using std::endl;
cannam@198: 
cannam@198: using Vamp::RealTime;
cannam@198: 
cannam@198: #include <cmath>
cannam@198: 
cannam@198: 
cannam@243: class FixedTempoEstimator::D
cannam@255: // this class just avoids us having to declare any data members in the header
cannam@243: {
cannam@243: public:
cannam@243:     D(float inputSampleRate);
cannam@243:     ~D();
cannam@243: 
cannam@243:     size_t getPreferredStepSize() const { return 64; }
cannam@243:     size_t getPreferredBlockSize() const { return 256; }
cannam@243: 
cannam@243:     ParameterList getParameterDescriptors() const;
cannam@243:     float getParameter(string id) const;
cannam@243:     void setParameter(string id, float value);
cannam@243: 
cannam@243:     OutputList getOutputDescriptors() const;
cannam@243: 
cannam@243:     bool initialise(size_t channels, size_t stepSize, size_t blockSize);
cannam@243:     void reset();
cannam@243:     FeatureSet process(const float *const *, RealTime);
cannam@243:     FeatureSet getRemainingFeatures();
cannam@243: 
cannam@243: private:
cannam@243:     void calculate();
cannam@243:     FeatureSet assembleFeatures();
cannam@243: 
cannam@243:     float lag2tempo(int);
cannam@243:     int tempo2lag(float);
cannam@243: 
cannam@243:     float m_inputSampleRate;
cannam@243:     size_t m_stepSize;
cannam@243:     size_t m_blockSize;
cannam@243: 
cannam@243:     float m_minbpm;
cannam@243:     float m_maxbpm;
cannam@243:     float m_maxdflen;
cannam@243: 
cannam@243:     float *m_priorMagnitudes;
cannam@243: 
cannam@243:     size_t m_dfsize;
cannam@243:     float *m_df;
cannam@243:     float *m_r;
cannam@243:     float *m_fr;
cannam@243:     float *m_t;
cannam@243:     size_t m_n;
cannam@243: 
cannam@243:     Vamp::RealTime m_start;
cannam@243:     Vamp::RealTime m_lasttime;
cannam@243: };
cannam@243: 
cannam@243: FixedTempoEstimator::D::D(float inputSampleRate) :
cannam@243:     m_inputSampleRate(inputSampleRate),
cannam@198:     m_stepSize(0),
cannam@198:     m_blockSize(0),
cannam@243:     m_minbpm(50),
cannam@243:     m_maxbpm(190),
cannam@243:     m_maxdflen(10),
cannam@198:     m_priorMagnitudes(0),
cannam@200:     m_df(0),
cannam@200:     m_r(0),
cannam@200:     m_fr(0),
cannam@204:     m_t(0),
cannam@200:     m_n(0)
cannam@198: {
cannam@198: }
cannam@198: 
cannam@243: FixedTempoEstimator::D::~D()
cannam@198: {
cannam@198:     delete[] m_priorMagnitudes;
cannam@198:     delete[] m_df;
cannam@200:     delete[] m_r;
cannam@200:     delete[] m_fr;
cannam@204:     delete[] m_t;
cannam@198: }
cannam@198: 
cannam@198: FixedTempoEstimator::ParameterList
cannam@243: FixedTempoEstimator::D::getParameterDescriptors() const
cannam@198: {
cannam@198:     ParameterList list;
cannam@243: 
cannam@243:     ParameterDescriptor d;
cannam@243:     d.identifier = "minbpm";
cannam@243:     d.name = "Minimum estimated tempo";
cannam@243:     d.description = "Minimum beat-per-minute value which the tempo estimator is able to return";
cannam@243:     d.unit = "bpm";
cannam@243:     d.minValue = 10;
cannam@243:     d.maxValue = 360;
cannam@243:     d.defaultValue = 50;
cannam@243:     d.isQuantized = false;
cannam@243:     list.push_back(d);
cannam@243: 
cannam@243:     d.identifier = "maxbpm";
cannam@243:     d.name = "Maximum estimated tempo";
cannam@243:     d.description = "Maximum beat-per-minute value which the tempo estimator is able to return";
cannam@243:     d.defaultValue = 190;
cannam@243:     list.push_back(d);
cannam@243: 
cannam@243:     d.identifier = "maxdflen";
cannam@243:     d.name = "Input duration to study";
cannam@243:     d.description = "Length of audio input, in seconds, which should be taken into account when estimating tempo.  There is no need to supply the plugin with any further input once this time has elapsed since the start of the audio.  The tempo estimator may use only the first part of this, up to eight times the slowest beat duration: increasing this value further than that is unlikely to improve results.";
cannam@243:     d.unit = "s";
cannam@243:     d.minValue = 2;
cannam@243:     d.maxValue = 40;
cannam@243:     d.defaultValue = 10;
cannam@243:     list.push_back(d);
cannam@243: 
cannam@198:     return list;
cannam@198: }
cannam@198: 
cannam@198: float
cannam@243: FixedTempoEstimator::D::getParameter(string id) const
cannam@198: {
cannam@243:     if (id == "minbpm") {
cannam@243:         return m_minbpm;
cannam@243:     } else if (id == "maxbpm") {
cannam@243:         return m_maxbpm;
cannam@243:     } else if (id == "maxdflen") {
cannam@243:         return m_maxdflen;
cannam@243:     }
cannam@198:     return 0.f;
cannam@198: }
cannam@198: 
cannam@198: void
cannam@243: FixedTempoEstimator::D::setParameter(string id, float value)
cannam@198: {
cannam@243:     if (id == "minbpm") {
cannam@243:         m_minbpm = value;
cannam@243:     } else if (id == "maxbpm") {
cannam@243:         m_maxbpm = value;
cannam@243:     } else if (id == "maxdflen") {
cannam@243:         m_maxdflen = value;
cannam@243:     }
cannam@198: }
cannam@198: 
cannam@200: static int TempoOutput = 0;
cannam@200: static int CandidatesOutput = 1;
cannam@200: static int DFOutput = 2;
cannam@200: static int ACFOutput = 3;
cannam@200: static int FilteredACFOutput = 4;
cannam@200: 
cannam@198: FixedTempoEstimator::OutputList
cannam@243: FixedTempoEstimator::D::getOutputDescriptors() const
cannam@198: {
cannam@198:     OutputList list;
cannam@198: 
cannam@198:     OutputDescriptor d;
cannam@198:     d.identifier = "tempo";
cannam@198:     d.name = "Tempo";
cannam@198:     d.description = "Estimated tempo";
cannam@198:     d.unit = "bpm";
cannam@198:     d.hasFixedBinCount = true;
cannam@198:     d.binCount = 1;
cannam@198:     d.hasKnownExtents = false;
cannam@198:     d.isQuantized = false;
cannam@198:     d.sampleType = OutputDescriptor::VariableSampleRate;
cannam@198:     d.sampleRate = m_inputSampleRate;
cannam@198:     d.hasDuration = true; // our returned tempo spans a certain range
cannam@198:     list.push_back(d);
cannam@198: 
cannam@200:     d.identifier = "candidates";
cannam@200:     d.name = "Tempo candidates";
cannam@200:     d.description = "Possible tempo estimates, one per bin with the most likely in the first bin";
cannam@200:     d.unit = "bpm";
cannam@200:     d.hasFixedBinCount = false;
cannam@200:     list.push_back(d);
cannam@200: 
cannam@198:     d.identifier = "detectionfunction";
cannam@198:     d.name = "Detection Function";
cannam@198:     d.description = "Onset detection function";
cannam@198:     d.unit = "";
cannam@198:     d.hasFixedBinCount = 1;
cannam@198:     d.binCount = 1;
cannam@198:     d.hasKnownExtents = true;
cannam@198:     d.minValue = 0.0;
cannam@198:     d.maxValue = 1.0;
cannam@198:     d.isQuantized = false;
cannam@198:     d.quantizeStep = 0.0;
cannam@198:     d.sampleType = OutputDescriptor::FixedSampleRate;
cannam@198:     if (m_stepSize) {
cannam@198:         d.sampleRate = m_inputSampleRate / m_stepSize;
cannam@198:     } else {
cannam@198:         d.sampleRate = m_inputSampleRate / (getPreferredBlockSize()/2);
cannam@198:     }
cannam@198:     d.hasDuration = false;
cannam@198:     list.push_back(d);
cannam@198: 
cannam@198:     d.identifier = "acf";
cannam@198:     d.name = "Autocorrelation Function";
cannam@198:     d.description = "Autocorrelation of onset detection function";
cannam@198:     d.hasKnownExtents = false;
cannam@201:     d.unit = "r";
cannam@198:     list.push_back(d);
cannam@198: 
cannam@198:     d.identifier = "filtered_acf";
cannam@198:     d.name = "Filtered Autocorrelation";
cannam@198:     d.description = "Filtered autocorrelation of onset detection function";
cannam@201:     d.unit = "r";
cannam@198:     list.push_back(d);
cannam@198: 
cannam@198:     return list;
cannam@198: }
cannam@198: 
cannam@243: bool
cannam@271: FixedTempoEstimator::D::initialise(size_t, size_t stepSize, size_t blockSize)
cannam@243: {
cannam@243:     m_stepSize = stepSize;
cannam@243:     m_blockSize = blockSize;
cannam@243: 
cannam@243:     float dfLengthSecs = m_maxdflen;
cannam@243:     m_dfsize = (dfLengthSecs * m_inputSampleRate) / m_stepSize;
cannam@243: 
cannam@243:     m_priorMagnitudes = new float[m_blockSize/2];
cannam@243:     m_df = new float[m_dfsize];
cannam@243: 
cannam@243:     for (size_t i = 0; i < m_blockSize/2; ++i) {
cannam@243:         m_priorMagnitudes[i] = 0.f;
cannam@243:     }
cannam@243:     for (size_t i = 0; i < m_dfsize; ++i) {
cannam@243:         m_df[i] = 0.f;
cannam@243:     }
cannam@243: 
cannam@243:     m_n = 0;
cannam@243: 
cannam@243:     return true;
cannam@243: }
cannam@243: 
cannam@243: void
cannam@243: FixedTempoEstimator::D::reset()
cannam@243: {
cannam@243:     if (!m_priorMagnitudes) return;
cannam@243: 
cannam@243:     for (size_t i = 0; i < m_blockSize/2; ++i) {
cannam@243:         m_priorMagnitudes[i] = 0.f;
cannam@243:     }
cannam@243:     for (size_t i = 0; i < m_dfsize; ++i) {
cannam@243:         m_df[i] = 0.f;
cannam@243:     }
cannam@243: 
cannam@243:     delete[] m_r;
cannam@243:     m_r = 0;
cannam@243: 
cannam@243:     delete[] m_fr; 
cannam@243:     m_fr = 0;
cannam@243: 
cannam@243:     delete[] m_t; 
cannam@243:     m_t = 0;
cannam@243: 
cannam@243:     m_n = 0;
cannam@243: 
cannam@243:     m_start = RealTime::zeroTime;
cannam@243:     m_lasttime = RealTime::zeroTime;
cannam@243: }
cannam@243: 
cannam@198: FixedTempoEstimator::FeatureSet
cannam@243: FixedTempoEstimator::D::process(const float *const *inputBuffers, RealTime ts)
cannam@198: {
cannam@198:     FeatureSet fs;
cannam@198: 
cannam@198:     if (m_stepSize == 0) {
cannam@198: 	cerr << "ERROR: FixedTempoEstimator::process: "
cannam@198: 	     << "FixedTempoEstimator has not been initialised"
cannam@198: 	     << endl;
cannam@198: 	return fs;
cannam@198:     }
cannam@198: 
cannam@198:     if (m_n == 0) m_start = ts;
cannam@198:     m_lasttime = ts;
cannam@198: 
cannam@198:     if (m_n == m_dfsize) {
cannam@255:         // If we have seen enough input, do the estimation and return
cannam@200:         calculate();
cannam@200:         fs = assembleFeatures();
cannam@198:         ++m_n;
cannam@198:         return fs;
cannam@198:     }
cannam@198: 
cannam@255:     // If we have seen more than enough, just discard and return!
cannam@198:     if (m_n > m_dfsize) return FeatureSet();
cannam@198: 
cannam@207:     float value = 0.f;
cannam@207: 
cannam@255:     // m_df will contain an onset detection function based on the rise
cannam@255:     // in overall power from one spectral frame to the next --
cannam@255:     // simplistic but reasonably effective for our purposes.
cannam@255: 
cannam@198:     for (size_t i = 1; i < m_blockSize/2; ++i) {
cannam@198: 
cannam@198:         float real = inputBuffers[0][i*2];
cannam@198:         float imag = inputBuffers[0][i*2 + 1];
cannam@198: 
cannam@198:         float sqrmag = real * real + imag * imag;
cannam@207:         value += fabsf(sqrmag - m_priorMagnitudes[i]);
cannam@198: 
cannam@198:         m_priorMagnitudes[i] = sqrmag;
cannam@198:     }
cannam@198: 
cannam@207:     m_df[m_n] = value;
cannam@207: 
cannam@198:     ++m_n;
cannam@198:     return fs;
cannam@243: }    
cannam@198: 
cannam@198: FixedTempoEstimator::FeatureSet
cannam@243: FixedTempoEstimator::D::getRemainingFeatures()
cannam@198: {
cannam@198:     FeatureSet fs;
cannam@198:     if (m_n > m_dfsize) return fs;
cannam@200:     calculate();
cannam@200:     fs = assembleFeatures();
cannam@198:     ++m_n;
cannam@198:     return fs;
cannam@198: }
cannam@198: 
cannam@198: float
cannam@243: FixedTempoEstimator::D::lag2tempo(int lag)
cannam@199: {
cannam@198:     return 60.f / ((lag * m_stepSize) / m_inputSampleRate);
cannam@198: }
cannam@198: 
cannam@207: int
cannam@243: FixedTempoEstimator::D::tempo2lag(float tempo)
cannam@207: {
cannam@207:     return ((60.f / tempo) * m_inputSampleRate) / m_stepSize;
cannam@207: }
cannam@207: 
cannam@200: void
cannam@243: FixedTempoEstimator::D::calculate()
cannam@200: {    
cannam@200:     if (m_r) {
cannam@207:         cerr << "FixedTempoEstimator::calculate: calculation already happened?" << endl;
cannam@200:         return;
cannam@200:     }
cannam@200: 
cannam@243:     if (m_n < m_dfsize / 9 &&
cannam@243:         m_n < (1.0 * m_inputSampleRate) / m_stepSize) { // 1 second
cannam@243:         cerr << "FixedTempoEstimator::calculate: Input is too short" << endl;
cannam@243:         return;
cannam@200:     }
cannam@200: 
cannam@255:     // This function takes m_df (the detection function array filled
cannam@255:     // out in process()) and calculates m_r (the raw autocorrelation)
cannam@255:     // and m_fr (the filtered autocorrelation from whose peaks tempo
cannam@255:     // estimates will be taken).
cannam@200: 
cannam@255:     int n = m_n; // length of actual df array (m_dfsize is the theoretical max)
cannam@255: 
cannam@255:     m_r  = new float[n/2]; // raw autocorrelation
cannam@255:     m_fr = new float[n/2]; // filtered autocorrelation
cannam@255:     m_t  = new float[n/2]; // averaged tempo estimate for each lag value
cannam@200: 
cannam@200:     for (int i = 0; i < n/2; ++i) {
cannam@255:         m_r[i]  = 0.f;
cannam@200:         m_fr[i] = 0.f;
cannam@255:         m_t[i]  = lag2tempo(i);
cannam@200:     }
cannam@200: 
cannam@255:     // Calculate the raw autocorrelation of the detection function
cannam@255: 
cannam@200:     for (int i = 0; i < n/2; ++i) {
cannam@200: 
cannam@271:         for (int j = i; j < n; ++j) {
cannam@200:             m_r[i] += m_df[j] * m_df[j - i];
cannam@200:         }
cannam@200: 
cannam@200:         m_r[i] /= n - i - 1;
cannam@200:     }
cannam@200: 
cannam@255:     // Filter the autocorrelation and average out the tempo estimates
cannam@255:     
cannam@246:     float related[] = { 0.5, 2, 4, 8 };
cannam@208: 
cannam@209:     for (int i = 1; i < n/2-1; ++i) {
cannam@204: 
cannam@209:         m_fr[i] = m_r[i];
cannam@204: 
cannam@200:         int div = 1;
cannam@200: 
cannam@215:         for (int j = 0; j < int(sizeof(related)/sizeof(related[0])); ++j) {
cannam@204: 
cannam@255:             // Check for an obvious peak at each metrically related lag
cannam@255: 
cannam@215:             int k0 = int(i * related[j] + 0.5);
cannam@209: 
cannam@215:             if (k0 >= 0 && k0 < int(n/2)) {
cannam@204: 
cannam@207:                 int kmax = 0, kmin = 0;
cannam@207:                 float kvmax = 0, kvmin = 0;
cannam@209:                 bool have = false;
cannam@204: 
cannam@209:                 for (int k = k0 - 1; k <= k0 + 1; ++k) {
cannam@204: 
cannam@209:                     if (k < 0 || k >= n/2) continue;
cannam@209: 
cannam@215:                     if (!have || (m_r[k] > kvmax)) { kmax = k; kvmax = m_r[k]; }
cannam@215:                     if (!have || (m_r[k] < kvmin)) { kmin = k; kvmin = m_r[k]; }
cannam@209:                     
cannam@209:                     have = true;
cannam@204:                 }
cannam@209:                 
cannam@255:                 // Boost the original lag according to the strongest
cannam@255:                 // value found close to this related lag
cannam@255: 
cannam@215:                 m_fr[i] += m_r[kmax] / 5;
cannam@209: 
cannam@209:                 if ((kmax == 0 || m_r[kmax] > m_r[kmax-1]) &&
cannam@209:                     (kmax == n/2-1 || m_r[kmax] > m_r[kmax+1]) &&
cannam@207:                     kvmax > kvmin * 1.05) {
cannam@255: 
cannam@255:                     // The strongest value close to the related lag is
cannam@255:                     // also a pretty good looking peak, so use it to
cannam@255:                     // improve our tempo estimate for the original lag
cannam@209:                     
cannam@207:                     m_t[i] = m_t[i] + lag2tempo(kmax) * related[j];
cannam@207:                     ++div;
cannam@207:                 }
cannam@204:             }
cannam@204:         }
cannam@209:         
cannam@204:         m_t[i] /= div;
cannam@204:         
cannam@255:         // Finally apply a primitive perceptual weighting (to prefer
cannam@255:         // tempi of around 120-130)
cannam@255: 
cannam@255:         float weight = 1.f - fabsf(128.f - lag2tempo(i)) * 0.005;
cannam@255:         if (weight < 0.f) weight = 0.f;
cannam@255:         weight = weight * weight * weight;
cannam@255: 
cannam@215:         m_fr[i] += m_fr[i] * (weight / 3);
cannam@207:     }
cannam@200: }
cannam@200:     
cannam@198: FixedTempoEstimator::FeatureSet
cannam@243: FixedTempoEstimator::D::assembleFeatures()
cannam@198: {
cannam@198:     FeatureSet fs;
cannam@255:     if (!m_r) return fs; // No autocorrelation: no results
cannam@200: 
cannam@198:     Feature feature;
cannam@198:     feature.hasTimestamp = true;
cannam@198:     feature.hasDuration = false;
cannam@198:     feature.label = "";
cannam@198:     feature.values.clear();
cannam@198:     feature.values.push_back(0.f);
cannam@198: 
cannam@200:     char buffer[40];
cannam@198: 
cannam@198:     int n = m_n;
cannam@198: 
cannam@198:     for (int i = 0; i < n; ++i) {
cannam@255: 
cannam@255:         // Return the detection function in the DF output
cannam@255: 
cannam@208:         feature.timestamp = m_start +
cannam@208:             RealTime::frame2RealTime(i * m_stepSize, m_inputSampleRate);
cannam@200:         feature.values[0] = m_df[i];
cannam@198:         feature.label = "";
cannam@200:         fs[DFOutput].push_back(feature);
cannam@198:     }
cannam@198: 
cannam@199:     for (int i = 1; i < n/2; ++i) {
cannam@255: 
cannam@255:         // Return the raw autocorrelation in the ACF output, each
cannam@255:         // value labelled according to its corresponding tempo
cannam@255: 
cannam@208:         feature.timestamp = m_start +
cannam@208:             RealTime::frame2RealTime(i * m_stepSize, m_inputSampleRate);
cannam@200:         feature.values[0] = m_r[i];
cannam@199:         sprintf(buffer, "%.1f bpm", lag2tempo(i));
cannam@200:         if (i == n/2-1) feature.label = "";
cannam@200:         else feature.label = buffer;
cannam@200:         fs[ACFOutput].push_back(feature);
cannam@198:     }
cannam@198: 
cannam@243:     float t0 = m_minbpm; // our minimum detected tempo
cannam@243:     float t1 = m_maxbpm; // our maximum detected tempo
cannam@216: 
cannam@207:     int p0 = tempo2lag(t1);
cannam@207:     int p1 = tempo2lag(t0);
cannam@198: 
cannam@200:     std::map<float, int> candidates;
cannam@198: 
cannam@271:     for (int i = p0; i <= p1 && i+1 < n/2; ++i) {
cannam@198: 
cannam@209:         if (m_fr[i] > m_fr[i-1] &&
cannam@209:             m_fr[i] > m_fr[i+1]) {
cannam@255: 
cannam@255:             // This is a peak in the filtered autocorrelation: stick
cannam@255:             // it into the map from filtered autocorrelation to lag
cannam@255:             // index -- this sorts our peaks by filtered acf value
cannam@255: 
cannam@209:             candidates[m_fr[i]] = i;
cannam@209:         }
cannam@198: 
cannam@255:         // Also return the filtered autocorrelation in its own output
cannam@255: 
cannam@208:         feature.timestamp = m_start +
cannam@208:             RealTime::frame2RealTime(i * m_stepSize, m_inputSampleRate);
cannam@200:         feature.values[0] = m_fr[i];
cannam@199:         sprintf(buffer, "%.1f bpm", lag2tempo(i));
cannam@200:         if (i == p1 || i == n/2-2) feature.label = "";
cannam@200:         else feature.label = buffer;
cannam@200:         fs[FilteredACFOutput].push_back(feature);
cannam@198:     }
cannam@198: 
cannam@200:     if (candidates.empty()) {
cannam@207:         cerr << "No tempo candidates!" << endl;
cannam@200:         return fs;
cannam@200:     }
cannam@198: 
cannam@198:     feature.hasTimestamp = true;
cannam@198:     feature.timestamp = m_start;
cannam@198:     
cannam@198:     feature.hasDuration = true;
cannam@198:     feature.duration = m_lasttime - m_start;
cannam@198: 
cannam@255:     // The map contains only peaks and is sorted by filtered acf
cannam@255:     // value, so the final element in it is our "best" tempo guess
cannam@255: 
cannam@200:     std::map<float, int>::const_iterator ci = candidates.end();
cannam@200:     --ci;
cannam@200:     int maxpi = ci->second;
cannam@198: 
cannam@204:     if (m_t[maxpi] > 0) {
cannam@255: 
cannam@255:         // This lag has an adjusted tempo from the averaging process:
cannam@255:         // use it
cannam@255: 
cannam@204:         feature.values[0] = m_t[maxpi];
cannam@255: 
cannam@204:     } else {
cannam@255: 
cannam@255:         // shouldn't happen -- it would imply that this high value was
cannam@255:         // not a peak!
cannam@255: 
cannam@204:         feature.values[0] = lag2tempo(maxpi);
cannam@207:         cerr << "WARNING: No stored tempo for index " << maxpi << endl;
cannam@204:     }
cannam@204: 
cannam@204:     sprintf(buffer, "%.1f bpm", feature.values[0]);
cannam@199:     feature.label = buffer;
cannam@199: 
cannam@255:     // Return the best tempo in the main output
cannam@255: 
cannam@200:     fs[TempoOutput].push_back(feature);
cannam@198: 
cannam@255:     // And return the other estimates (up to the arbitrarily chosen
cannam@255:     // number of 10 of them) in the candidates output
cannam@255: 
cannam@200:     feature.values.clear();
cannam@200:     feature.label = "";
cannam@200: 
cannam@255:     while (feature.values.size() < 10) {
cannam@207:         if (m_t[ci->second] > 0) {
cannam@207:             feature.values.push_back(m_t[ci->second]);
cannam@207:         } else {
cannam@207:             feature.values.push_back(lag2tempo(ci->second));
cannam@207:         }
cannam@200:         if (ci == candidates.begin()) break;
cannam@200:         --ci;
cannam@200:     }
cannam@200: 
cannam@200:     fs[CandidatesOutput].push_back(feature);
cannam@200:     
cannam@198:     return fs;
cannam@198: }
cannam@243: 
cannam@243:     
cannam@243: 
cannam@243: FixedTempoEstimator::FixedTempoEstimator(float inputSampleRate) :
cannam@243:     Plugin(inputSampleRate),
cannam@243:     m_d(new D(inputSampleRate))
cannam@243: {
cannam@243: }
cannam@243: 
cannam@243: FixedTempoEstimator::~FixedTempoEstimator()
cannam@243: {
cannam@271:     delete m_d;
cannam@243: }
cannam@243: 
cannam@243: string
cannam@243: FixedTempoEstimator::getIdentifier() const
cannam@243: {
cannam@243:     return "fixedtempo";
cannam@243: }
cannam@243: 
cannam@243: string
cannam@243: FixedTempoEstimator::getName() const
cannam@243: {
cannam@243:     return "Simple Fixed Tempo Estimator";
cannam@243: }
cannam@243: 
cannam@243: string
cannam@243: FixedTempoEstimator::getDescription() const
cannam@243: {
cannam@243:     return "Study a short section of audio and estimate its tempo, assuming the tempo is constant";
cannam@243: }
cannam@243: 
cannam@243: string
cannam@243: FixedTempoEstimator::getMaker() const
cannam@243: {
cannam@243:     return "Vamp SDK Example Plugins";
cannam@243: }
cannam@243: 
cannam@243: int
cannam@243: FixedTempoEstimator::getPluginVersion() const
cannam@243: {
cannam@243:     return 1;
cannam@243: }
cannam@243: 
cannam@243: string
cannam@243: FixedTempoEstimator::getCopyright() const
cannam@243: {
cannam@243:     return "Code copyright 2008 Queen Mary, University of London.  Freely redistributable (BSD license)";
cannam@243: }
cannam@243: 
cannam@243: size_t
cannam@243: FixedTempoEstimator::getPreferredStepSize() const
cannam@243: {
cannam@243:     return m_d->getPreferredStepSize();
cannam@243: }
cannam@243: 
cannam@243: size_t
cannam@243: FixedTempoEstimator::getPreferredBlockSize() const
cannam@243: {
cannam@243:     return m_d->getPreferredBlockSize();
cannam@243: }
cannam@243: 
cannam@243: bool
cannam@243: FixedTempoEstimator::initialise(size_t channels, size_t stepSize, size_t blockSize)
cannam@243: {
cannam@243:     if (channels < getMinChannelCount() ||
cannam@243: 	channels > getMaxChannelCount()) return false;
cannam@243: 
cannam@243:     return m_d->initialise(channels, stepSize, blockSize);
cannam@243: }
cannam@243: 
cannam@243: void
cannam@243: FixedTempoEstimator::reset()
cannam@243: {
cannam@243:     return m_d->reset();
cannam@243: }
cannam@243: 
cannam@243: FixedTempoEstimator::ParameterList
cannam@243: FixedTempoEstimator::getParameterDescriptors() const
cannam@243: {
cannam@243:     return m_d->getParameterDescriptors();
cannam@243: }
cannam@243: 
cannam@243: float
cannam@243: FixedTempoEstimator::getParameter(std::string id) const
cannam@243: {
cannam@243:     return m_d->getParameter(id);
cannam@243: }
cannam@243: 
cannam@243: void
cannam@243: FixedTempoEstimator::setParameter(std::string id, float value)
cannam@243: {
cannam@243:     m_d->setParameter(id, value);
cannam@243: }
cannam@243: 
cannam@243: FixedTempoEstimator::OutputList
cannam@243: FixedTempoEstimator::getOutputDescriptors() const
cannam@243: {
cannam@243:     return m_d->getOutputDescriptors();
cannam@243: }
cannam@243: 
cannam@243: FixedTempoEstimator::FeatureSet
cannam@243: FixedTempoEstimator::process(const float *const *inputBuffers, RealTime ts)
cannam@243: {
cannam@243:     return m_d->process(inputBuffers, ts);
cannam@243: }
cannam@243: 
cannam@243: FixedTempoEstimator::FeatureSet
cannam@243: FixedTempoEstimator::getRemainingFeatures()
cannam@243: {
cannam@243:     return m_d->getRemainingFeatures();
cannam@243: }