annotate dsp/segmentation/ClusterMeltSegmenter.h @ 251:c3600d3cfe5c

* Add timbral (MFCC) feature option to segmenter
author Chris Cannam <c.cannam@qmul.ac.uk>
date Thu, 10 Jan 2008 16:41:33 +0000
parents 18a0dffa5c1a
children 6cb2b3cd5356
rev   line source
c@249 1 /* -*- c-basic-offset: 4 indent-tabs-mode: nil -*- vi:set ts=8 sts=4 sw=4: */
c@249 2
c@243 3 /*
c@249 4 * ClusterMeltSegmenter.h
c@243 5 *
c@249 6 * Created by Mark Levy on 23/03/2006.
c@249 7 * Copyright 2006 Centre for Digital Music, Queen Mary, University of London.
c@249 8 * All rights reserved.
c@243 9 */
c@243 10
c@243 11 #include <vector>
c@243 12
c@243 13 #include "segment.h"
c@243 14 #include "Segmenter.h"
c@245 15 #include "hmm/hmm.h"
c@245 16 #include "base/Window.h"
c@243 17
c@243 18 using std::vector;
c@243 19
c@249 20 class Decimator;
c@249 21 class ConstantQ;
c@251 22 class MFCC;
c@249 23
c@249 24 class ClusterMeltSegmenterParams
c@249 25 // defaults are sensible for 11025Hz with 0.2 second hopsize
c@243 26 {
c@243 27 public:
c@249 28 ClusterMeltSegmenterParams() :
c@249 29 featureType(FEATURE_TYPE_CONSTQ),
c@249 30 hopSize(0.2),
c@249 31 windowSize(0.6),
c@249 32 fmin(62),
c@249 33 fmax(16000),
c@249 34 nbins(8),
c@249 35 ncomponents(20),
c@249 36 nHMMStates(40),
c@249 37 nclusters(10),
c@249 38 histogramLength(15),
c@249 39 neighbourhoodLimit(20) { }
c@249 40 feature_types featureType;
c@249 41 double hopSize; // in secs
c@249 42 double windowSize; // in secs
c@249 43 int fmin;
c@249 44 int fmax;
c@249 45 int nbins;
c@249 46 int ncomponents;
c@249 47 int nHMMStates;
c@249 48 int nclusters;
c@249 49 int histogramLength;
c@249 50 int neighbourhoodLimit;
c@243 51 };
c@243 52
c@243 53 class ClusterMeltSegmenter : public Segmenter
c@243 54 {
c@243 55 public:
c@249 56 ClusterMeltSegmenter(ClusterMeltSegmenterParams params);
c@249 57 virtual ~ClusterMeltSegmenter();
c@249 58 virtual void initialise(int samplerate);
c@249 59 virtual int getWindowsize();
c@249 60 virtual int getHopsize();
c@249 61 virtual void extractFeatures(const double* samples, int nsamples);
c@249 62 void setFeatures(const vector<vector<double> >& f); // provide the features yourself
c@249 63 virtual void segment(); // segment into default number of segment-types
c@249 64 void segment(int m); // segment into m segment-types
c@249 65 int getNSegmentTypes() { return nclusters; }
c@249 66
c@243 67 protected:
c@249 68 void makeSegmentation(int* q, int len);
c@243 69
c@251 70 void extractFeaturesConstQ(const double *, int);
c@251 71 void extractFeaturesMFCC(const double *, int);
c@251 72
c@249 73 Window<double> *window;
c@251 74 ConstantQ* constq;
c@251 75 MFCC* mfcc;
c@249 76 model_t* model; // the HMM
c@249 77 int* q; // the decoded HMM state sequence
c@249 78 vector<vector<double> > histograms;
c@249 79
c@249 80 feature_types featureType;
c@249 81 double hopSize; // in seconds
c@249 82 double windowSize; // in seconds
c@249 83
c@249 84 // constant-Q parameters
c@249 85 int fmin;
c@249 86 int fmax;
c@249 87 int nbins;
c@249 88 int ncoeff;
c@249 89
c@249 90 // PCA parameters
c@249 91 int ncomponents;
c@249 92
c@249 93 // HMM parameters
c@249 94 int nHMMStates;
c@249 95
c@249 96 // clustering parameters
c@249 97 int nclusters;
c@249 98 int histogramLength;
c@249 99 int neighbourhoodLimit;
c@249 100
c@249 101 Decimator *decimator;
c@243 102 };