Chris@3: /* -*- c-basic-offset: 4 indent-tabs-mode: nil -*-  vi:set ts=8 sts=4 sw=4: */
Chris@3: /*
Chris@31:     This file is Copyright (c) 2012 Chris Cannam
Chris@31:   
Chris@3:     Permission is hereby granted, free of charge, to any person
Chris@3:     obtaining a copy of this software and associated documentation
Chris@3:     files (the "Software"), to deal in the Software without
Chris@3:     restriction, including without limitation the rights to use, copy,
Chris@3:     modify, merge, publish, distribute, sublicense, and/or sell copies
Chris@3:     of the Software, and to permit persons to whom the Software is
Chris@3:     furnished to do so, subject to the following conditions:
Chris@3: 
Chris@3:     The above copyright notice and this permission notice shall be
Chris@3:     included in all copies or substantial portions of the Software.
Chris@3: 
Chris@3:     THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
Chris@3:     EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
Chris@3:     MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
Chris@3:     NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR
Chris@3:     ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF
Chris@3:     CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
Chris@3:     WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
Chris@3: */
Chris@3: 
Chris@31: #include "CepstralPitchTracker.h"
Chris@51: #include "Cepstrum.h"
Chris@47: #include "MeanFilter.h"
Chris@50: #include "PeakInterpolator.h"
Chris@56: #include "AgentFeeder.h"
Chris@3: 
Chris@26: #include "vamp-sdk/FFT.h"
Chris@26: 
Chris@3: #include <vector>
Chris@3: #include <algorithm>
Chris@3: 
Chris@3: #include <cstdio>
Chris@3: #include <cmath>
Chris@3: #include <complex>
Chris@3: 
Chris@3: using std::string;
Chris@7: using std::vector;
Chris@16: using Vamp::RealTime;
Chris@7: 
Chris@16: 
Chris@31: CepstralPitchTracker::CepstralPitchTracker(float inputSampleRate) :
Chris@3:     Plugin(inputSampleRate),
Chris@3:     m_channels(0),
Chris@3:     m_stepSize(256),
Chris@3:     m_blockSize(1024),
Chris@3:     m_fmin(50),
Chris@25:     m_fmax(900),
Chris@18:     m_vflen(1),
Chris@3:     m_binFrom(0),
Chris@3:     m_binTo(0),
Chris@56:     m_bins(0),
Chris@57:     m_nAccepted(0),
Chris@56:     m_feeder(0)
Chris@3: {
Chris@3: }
Chris@3: 
Chris@31: CepstralPitchTracker::~CepstralPitchTracker()
Chris@3: {
Chris@56:     delete m_feeder;
Chris@3: }
Chris@3: 
Chris@3: string
Chris@31: CepstralPitchTracker::getIdentifier() const
Chris@3: {
Chris@39:     return "cepstral-pitchtracker";
Chris@3: }
Chris@3: 
Chris@3: string
Chris@31: CepstralPitchTracker::getName() const
Chris@3: {
Chris@39:     return "Cepstral Pitch Tracker";
Chris@3: }
Chris@3: 
Chris@3: string
Chris@31: CepstralPitchTracker::getDescription() const
Chris@3: {
Chris@3:     return "Estimate f0 of monophonic material using a cepstrum method.";
Chris@3: }
Chris@3: 
Chris@3: string
Chris@31: CepstralPitchTracker::getMaker() const
Chris@3: {
Chris@3:     return "Chris Cannam";
Chris@3: }
Chris@3: 
Chris@3: int
Chris@31: CepstralPitchTracker::getPluginVersion() const
Chris@3: {
Chris@3:     // Increment this each time you release a version that behaves
Chris@3:     // differently from the previous one
Chris@3:     return 1;
Chris@3: }
Chris@3: 
Chris@3: string
Chris@31: CepstralPitchTracker::getCopyright() const
Chris@3: {
Chris@3:     return "Freely redistributable (BSD license)";
Chris@3: }
Chris@3: 
Chris@31: CepstralPitchTracker::InputDomain
Chris@31: CepstralPitchTracker::getInputDomain() const
Chris@3: {
Chris@3:     return FrequencyDomain;
Chris@3: }
Chris@3: 
Chris@3: size_t
Chris@31: CepstralPitchTracker::getPreferredBlockSize() const
Chris@3: {
Chris@3:     return 1024;
Chris@3: }
Chris@3: 
Chris@3: size_t 
Chris@31: CepstralPitchTracker::getPreferredStepSize() const
Chris@3: {
Chris@3:     return 256;
Chris@3: }
Chris@3: 
Chris@3: size_t
Chris@31: CepstralPitchTracker::getMinChannelCount() const
Chris@3: {
Chris@3:     return 1;
Chris@3: }
Chris@3: 
Chris@3: size_t
Chris@31: CepstralPitchTracker::getMaxChannelCount() const
Chris@3: {
Chris@3:     return 1;
Chris@3: }
Chris@3: 
Chris@31: CepstralPitchTracker::ParameterList
Chris@31: CepstralPitchTracker::getParameterDescriptors() const
Chris@3: {
Chris@3:     ParameterList list;
Chris@3:     return list;
Chris@3: }
Chris@3: 
Chris@3: float
Chris@31: CepstralPitchTracker::getParameter(string identifier) const
Chris@3: {
Chris@3:     return 0.f;
Chris@3: }
Chris@3: 
Chris@3: void
Chris@31: CepstralPitchTracker::setParameter(string identifier, float value) 
Chris@3: {
Chris@3: }
Chris@3: 
Chris@31: CepstralPitchTracker::ProgramList
Chris@31: CepstralPitchTracker::getPrograms() const
Chris@3: {
Chris@3:     ProgramList list;
Chris@3:     return list;
Chris@3: }
Chris@3: 
Chris@3: string
Chris@31: CepstralPitchTracker::getCurrentProgram() const
Chris@3: {
Chris@3:     return ""; // no programs
Chris@3: }
Chris@3: 
Chris@3: void
Chris@31: CepstralPitchTracker::selectProgram(string name)
Chris@3: {
Chris@3: }
Chris@3: 
Chris@31: CepstralPitchTracker::OutputList
Chris@31: CepstralPitchTracker::getOutputDescriptors() const
Chris@3: {
Chris@3:     OutputList outputs;
Chris@3: 
Chris@3:     OutputDescriptor d;
Chris@3: 
Chris@3:     d.identifier = "f0";
Chris@3:     d.name = "Estimated f0";
Chris@3:     d.description = "Estimated fundamental frequency";
Chris@3:     d.unit = "Hz";
Chris@3:     d.hasFixedBinCount = true;
Chris@3:     d.binCount = 1;
Chris@3:     d.hasKnownExtents = true;
Chris@3:     d.minValue = m_fmin;
Chris@3:     d.maxValue = m_fmax;
Chris@3:     d.isQuantized = false;
Chris@3:     d.sampleType = OutputDescriptor::FixedSampleRate;
Chris@3:     d.sampleRate = (m_inputSampleRate / m_stepSize);
Chris@3:     d.hasDuration = false;
Chris@3:     outputs.push_back(d);
Chris@3: 
Chris@16:     d.identifier = "notes";
Chris@16:     d.name = "Notes";
Chris@16:     d.description = "Derived fixed-pitch note frequencies";
Chris@16:     d.unit = "Hz";
Chris@16:     d.hasFixedBinCount = true;
Chris@16:     d.binCount = 1;
Chris@16:     d.hasKnownExtents = true;
Chris@16:     d.minValue = m_fmin;
Chris@16:     d.maxValue = m_fmax;
Chris@16:     d.isQuantized = false;
Chris@16:     d.sampleType = OutputDescriptor::FixedSampleRate;
Chris@16:     d.sampleRate = (m_inputSampleRate / m_stepSize);
Chris@16:     d.hasDuration = true;
Chris@16:     outputs.push_back(d);
Chris@16: 
Chris@3:     return outputs;
Chris@3: }
Chris@3: 
Chris@3: bool
Chris@31: CepstralPitchTracker::initialise(size_t channels, size_t stepSize, size_t blockSize)
Chris@3: {
Chris@3:     if (channels < getMinChannelCount() ||
Chris@3: 	channels > getMaxChannelCount()) return false;
Chris@3: 
Chris@31: //    std::cerr << "CepstralPitchTracker::initialise: channels = " << channels
Chris@3: //	      << ", stepSize = " << stepSize << ", blockSize = " << blockSize
Chris@3: //	      << std::endl;
Chris@3: 
Chris@3:     m_channels = channels;
Chris@3:     m_stepSize = stepSize;
Chris@3:     m_blockSize = blockSize;
Chris@3: 
Chris@3:     m_binFrom = int(m_inputSampleRate / m_fmax);
Chris@3:     m_binTo = int(m_inputSampleRate / m_fmin); 
Chris@3: 
Chris@3:     if (m_binTo >= (int)m_blockSize / 2) {
Chris@3:         m_binTo = m_blockSize / 2 - 1;
Chris@3:     }
Chris@56:     if (m_binFrom >= m_binTo) {
Chris@56:         // shouldn't happen except for degenerate samplerate / blocksize combos
Chris@56:         m_binFrom = m_binTo - 1;
Chris@56:     }
Chris@3: 
Chris@3:     m_bins = (m_binTo - m_binFrom) + 1;
Chris@3: 
Chris@3:     reset();
Chris@3: 
Chris@3:     return true;
Chris@3: }
Chris@3: 
Chris@3: void
Chris@31: CepstralPitchTracker::reset()
Chris@3: {
Chris@56:     delete m_feeder;
Chris@56:     m_feeder = new AgentFeeder();
Chris@57:     m_nAccepted = 0;
Chris@3: }
Chris@3: 
Chris@3: void
Chris@35: CepstralPitchTracker::addFeaturesFrom(NoteHypothesis h, FeatureSet &fs)
Chris@30: {
Chris@35:     NoteHypothesis::Estimates es = h.getAcceptedEstimates();
Chris@30: 
Chris@35:     for (int i = 0; i < (int)es.size(); ++i) {
Chris@30: 	Feature f;
Chris@30: 	f.hasTimestamp = true;
Chris@30: 	f.timestamp = es[i].time;
Chris@30: 	f.values.push_back(es[i].freq);
Chris@30: 	fs[0].push_back(f);
Chris@30:     }
Chris@30: 
Chris@30:     Feature nf;
Chris@30:     nf.hasTimestamp = true;
Chris@30:     nf.hasDuration = true;
Chris@35:     NoteHypothesis::Note n = h.getAveragedNote();
Chris@30:     nf.timestamp = n.time;
Chris@30:     nf.duration = n.duration;
Chris@30:     nf.values.push_back(n.freq);
Chris@30:     fs[1].push_back(nf);
Chris@30: }
Chris@30: 
Chris@57: void
Chris@57: CepstralPitchTracker::addNewFeatures(FeatureSet &fs)
Chris@57: {
Chris@57:     int n = m_feeder->getAcceptedHypotheses().size();
Chris@57:     if (n == m_nAccepted) return;
Chris@57: 
Chris@57:     AgentFeeder::Hypotheses accepted = m_feeder->getAcceptedHypotheses();
Chris@57: 
Chris@57:     for (int i = m_nAccepted; i < n; ++i) {
Chris@57:         addFeaturesFrom(accepted[i], fs);
Chris@57:     }
Chris@57: 
Chris@57:     m_nAccepted = n;
Chris@57: }
Chris@57: 
Chris@31: CepstralPitchTracker::FeatureSet
Chris@31: CepstralPitchTracker::process(const float *const *inputBuffers, RealTime timestamp)
Chris@3: {
Chris@51:     double *rawcep = new double[m_blockSize];
Chris@51:     double magmean = Cepstrum(m_blockSize).process(inputBuffers[0], rawcep);
Chris@3: 
Chris@3:     int n = m_bins;
Chris@3:     double *data = new double[n];
Chris@51:     MeanFilter(m_vflen).filterSubsequence
Chris@51:         (rawcep, data, m_blockSize, n, m_binFrom);
Chris@51: 
Chris@3:     delete[] rawcep;
Chris@3: 
Chris@3:     double maxval = 0.0;
Chris@6:     int maxbin = -1;
Chris@3: 
Chris@3:     for (int i = 0; i < n; ++i) {
Chris@3:         if (data[i] > maxval) {
Chris@3:             maxval = data[i];
Chris@3:             maxbin = i;
Chris@3:         }
Chris@3:     }
Chris@3: 
Chris@15:     if (maxbin < 0) {
Chris@15:         delete[] data;
Chris@57:         return FeatureSet();
Chris@15:     }
Chris@15: 
Chris@15:     double nextPeakVal = 0.0;
Chris@15:     for (int i = 1; i+1 < n; ++i) {
Chris@15:         if (data[i] > data[i-1] &&
Chris@15:             data[i] > data[i+1] &&
Chris@15:             i != maxbin &&
Chris@15:             data[i] > nextPeakVal) {
Chris@15:             nextPeakVal = data[i];
Chris@15:         }
Chris@15:     }
Chris@8: 
Chris@50:     PeakInterpolator pi;
Chris@50:     double cimax = pi.findPeakLocation(data, m_bins, maxbin);
Chris@18:     double peakfreq = m_inputSampleRate / (cimax + m_binFrom);
Chris@15: 
Chris@15:     double confidence = 0.0;
Chris@51:     double threshold = 0.1; // for magmean
Chris@51: 
Chris@15:     if (nextPeakVal != 0.0) {
Chris@27:         confidence = (maxval - nextPeakVal) * 10.0;
Chris@25:         if (magmean < threshold) confidence = 0.0;
Chris@15:     }
Chris@15: 
Chris@57:     delete[] data;
Chris@57: 
Chris@35:     NoteHypothesis::Estimate e;
Chris@8:     e.freq = peakfreq;
Chris@8:     e.time = timestamp;
Chris@15:     e.confidence = confidence;
Chris@8: 
Chris@56:     m_feeder->feed(e);
Chris@14: 
Chris@57:     FeatureSet fs;
Chris@57:     addNewFeatures(fs);
Chris@3:     return fs;
Chris@3: }
Chris@3: 
Chris@31: CepstralPitchTracker::FeatureSet
Chris@31: CepstralPitchTracker::getRemainingFeatures()
Chris@3: {
Chris@56:     m_feeder->finish();
Chris@56: 
Chris@3:     FeatureSet fs;
Chris@57:     addNewFeatures(fs);
Chris@3:     return fs;
Chris@3: }