annotate dsp/tempotracking/DownBeat.h @ 309:d5014ab8b0e5

* Add GPL and README; some tidying
author Chris Cannam <c.cannam@qmul.ac.uk>
date Mon, 13 Dec 2010 14:55:28 +0000
parents befe5aa6b450
children 915365c58758
rev   line source
c@279 1 /* -*- c-basic-offset: 4 indent-tabs-mode: nil -*- vi:set ts=8 sts=4 sw=4: */
c@279 2
c@279 3 /*
c@279 4 QM DSP Library
c@279 5
c@279 6 Centre for Digital Music, Queen Mary, University of London.
c@279 7 This file copyright 2008-2009 Matthew Davies and QMUL.
c@309 8
c@309 9 This program is free software; you can redistribute it and/or
c@309 10 modify it under the terms of the GNU General Public License as
c@309 11 published by the Free Software Foundation; either version 2 of the
c@309 12 License, or (at your option) any later version. See the file
c@309 13 COPYING included with this distribution for more information.
c@279 14 */
c@279 15
c@279 16 #ifndef DOWNBEAT_H
c@279 17 #define DOWNBEAT_H
c@279 18
c@279 19 #include <vector>
c@279 20
c@279 21 #include "dsp/rateconversion/Decimator.h"
c@279 22
c@279 23 using std::vector;
c@279 24
c@289 25 class FFTReal;
c@289 26
c@279 27 /**
c@279 28 * This class takes an input audio signal and a sequence of beat
c@279 29 * locations (calculated e.g. by TempoTrackV2) and estimates which of
c@279 30 * the beat locations are downbeats (first beat of the bar).
c@279 31 *
c@279 32 * The input audio signal is expected to have been downsampled to a
c@279 33 * very low sampling rate (e.g. 2700Hz). A utility function for
c@279 34 * downsampling and buffering incoming block-by-block audio is
c@279 35 * provided.
c@279 36 */
c@279 37 class DownBeat
c@279 38 {
c@279 39 public:
c@279 40 /**
c@279 41 * Construct a downbeat locator that will operate on audio at the
c@279 42 * downsampled by the given decimation factor from the given
c@279 43 * original sample rate, plus beats extracted from the same audio
c@279 44 * at the given original sample rate with the given frame
c@279 45 * increment.
c@279 46 *
c@279 47 * decimationFactor must be a power of two no greater than 64, and
c@279 48 * dfIncrement must be a multiple of decimationFactor.
c@279 49 */
c@279 50 DownBeat(float originalSampleRate,
c@279 51 size_t decimationFactor,
c@279 52 size_t dfIncrement);
c@279 53 ~DownBeat();
c@279 54
c@280 55 void setBeatsPerBar(int bpb);
c@280 56
c@279 57 /**
c@279 58 * Estimate which beats are down-beats.
c@279 59 *
c@279 60 * audio contains the input audio stream after downsampling, and
c@279 61 * audioLength contains the number of samples in this downsampled
c@279 62 * stream.
c@279 63 *
c@279 64 * beats contains a series of beat positions expressed in
c@279 65 * multiples of the df increment at the audio's original sample
c@279 66 * rate, as described to the constructor.
c@279 67 *
c@279 68 * The returned downbeat array contains a series of indices to the
c@279 69 * beats array.
c@279 70 */
c@280 71 void findDownBeats(const float *audio, // downsampled
c@279 72 size_t audioLength, // after downsampling
c@279 73 const vector<double> &beats,
c@279 74 vector<int> &downbeats);
c@281 75
c@281 76 /**
c@281 77 * Return the beat spectral difference function. This is
c@281 78 * calculated during findDownBeats, so this function can only be
c@281 79 * meaningfully called after that has completed. The returned
c@281 80 * vector contains one value for each of the beat times passed in
c@281 81 * to findDownBeats, less one. Each value contains the spectral
c@281 82 * difference between region prior to the beat's nominal position
c@281 83 * and the region following it.
c@281 84 */
c@281 85 void getBeatSD(vector<double> &beatsd) const;
c@279 86
c@279 87 /**
c@279 88 * For your downsampling convenience: call this function
c@279 89 * repeatedly with input audio blocks containing dfIncrement
c@279 90 * samples at the original sample rate, to decimate them to the
c@279 91 * downsampled rate and buffer them within the DownBeat class.
c@279 92 *
c@279 93 * Call getBufferedAudio() to retrieve the results after all
c@279 94 * blocks have been processed.
c@279 95 */
c@280 96 void pushAudioBlock(const float *audio);
c@279 97
c@279 98 /**
c@279 99 * Retrieve the accumulated audio produced by pushAudioBlock calls.
c@279 100 */
c@280 101 const float *getBufferedAudio(size_t &length) const;
c@280 102
c@280 103 /**
c@280 104 * Clear any buffered downsampled audio data.
c@280 105 */
c@280 106 void resetAudioBuffer();
c@279 107
c@279 108 private:
c@279 109 typedef vector<int> i_vec_t;
c@279 110 typedef vector<vector<int> > i_mat_t;
c@279 111 typedef vector<double> d_vec_t;
c@279 112 typedef vector<vector<double> > d_mat_t;
c@279 113
c@279 114 void makeDecimators();
c@279 115 double measureSpecDiff(d_vec_t oldspec, d_vec_t newspec);
c@279 116
c@280 117 int m_bpb;
c@279 118 float m_rate;
c@279 119 size_t m_factor;
c@279 120 size_t m_increment;
c@279 121 Decimator *m_decimator1;
c@279 122 Decimator *m_decimator2;
c@280 123 float *m_buffer;
c@280 124 float *m_decbuf;
c@279 125 size_t m_bufsiz;
c@279 126 size_t m_buffill;
c@279 127 size_t m_beatframesize;
c@279 128 double *m_beatframe;
c@289 129 FFTReal *m_fft;
c@279 130 double *m_fftRealOut;
c@279 131 double *m_fftImagOut;
c@281 132 d_vec_t m_beatsd;
c@279 133 };
c@279 134
c@279 135 #endif