annotate audioio/PhaseVocoderTimeStretcher.h @ 202:738968d96e19

...
author Chris Cannam
date Wed, 24 Oct 2007 16:00:30 +0000
parents f3516e520652
children
rev   line source
Chris@0 1 /* -*- c-basic-offset: 4 indent-tabs-mode: nil -*- vi:set ts=8 sts=4 sw=4: */
Chris@0 2
Chris@0 3 /*
Chris@0 4 Sonic Visualiser
Chris@0 5 An audio file viewer and annotation editor.
Chris@0 6 Centre for Digital Music, Queen Mary, University of London.
Chris@77 7 This file copyright 2006 Chris Cannam and QMUL.
Chris@0 8
Chris@0 9 This program is free software; you can redistribute it and/or
Chris@0 10 modify it under the terms of the GNU General Public License as
Chris@0 11 published by the Free Software Foundation; either version 2 of the
Chris@0 12 License, or (at your option) any later version. See the file
Chris@0 13 COPYING included with this distribution for more information.
Chris@0 14 */
Chris@0 15
Chris@14 16 #ifndef _PHASE_VOCODER_TIME_STRETCHER_H_
Chris@14 17 #define _PHASE_VOCODER_TIME_STRETCHER_H_
Chris@0 18
Chris@0 19 #include "base/Window.h"
Chris@0 20 #include "base/RingBuffer.h"
Chris@0 21
Chris@100 22 #include "data/fft/FFTapi.h"
Chris@0 23
Chris@25 24 #include <QMutex>
Chris@25 25
Chris@0 26 /**
Chris@14 27 * A time stretcher that alters the performance speed of audio,
Chris@22 28 * preserving pitch.
Chris@0 29 *
Chris@22 30 * This is based on the straightforward phase vocoder with phase
Chris@22 31 * unwrapping (as in e.g. the DAFX book pp275-), with optional
Chris@22 32 * percussive transient detection to avoid smearing percussive notes
Chris@22 33 * and resynchronise phases, and adding a stream API for real-time
Chris@22 34 * use. Principles and methods from Chris Duxbury, AES 2002 and 2004
Chris@22 35 * thesis; Emmanuel Ravelli, DAFX 2005; Dan Barry, ISSC 2005 on
Chris@22 36 * percussion detection; code by Chris Cannam.
Chris@0 37 */
Chris@0 38
Chris@14 39 class PhaseVocoderTimeStretcher
Chris@0 40 {
Chris@0 41 public:
Chris@22 42 PhaseVocoderTimeStretcher(size_t sampleRate,
Chris@22 43 size_t channels,
Chris@16 44 float ratio,
Chris@16 45 bool sharpen,
Chris@31 46 size_t maxOutputBlockSize);
Chris@14 47 virtual ~PhaseVocoderTimeStretcher();
Chris@0 48
Chris@12 49 /**
Chris@16 50 * Return the number of samples that would need to be added via
Chris@16 51 * putInput in order to provoke the time stretcher into doing some
Chris@16 52 * time stretching and making more output samples available.
Chris@31 53 * This will be an estimate, if transient sharpening is on; the
Chris@31 54 * caller may need to do the put/get/test cycle more than once.
Chris@16 55 */
Chris@16 56 size_t getRequiredInputSamples() const;
Chris@16 57
Chris@16 58 /**
Chris@16 59 * Put (and possibly process) a given number of input samples.
Chris@31 60 * Number should usually equal the value returned from
Chris@31 61 * getRequiredInputSamples().
Chris@16 62 */
Chris@16 63 void putInput(float **input, size_t samples);
Chris@16 64
Chris@23 65 /**
Chris@23 66 * Get the number of processed samples ready for reading.
Chris@23 67 */
Chris@16 68 size_t getAvailableOutputSamples() const;
Chris@16 69
Chris@23 70 /**
Chris@23 71 * Get some processed samples.
Chris@23 72 */
Chris@16 73 void getOutput(float **output, size_t samples);
Chris@16 74
Chris@16 75 //!!! and reset?
Chris@0 76
Chris@0 77 /**
Chris@25 78 * Change the time stretch ratio.
Chris@25 79 */
Chris@25 80 void setRatio(float ratio);
Chris@25 81
Chris@25 82 /**
Chris@15 83 * Get the hop size for input.
Chris@0 84 */
Chris@0 85 size_t getInputIncrement() const { return m_n1; }
Chris@0 86
Chris@0 87 /**
Chris@15 88 * Get the hop size for output.
Chris@15 89 */
Chris@16 90 size_t getOutputIncrement() const { return m_n2; }
Chris@15 91
Chris@15 92 /**
Chris@15 93 * Get the window size for FFT processing.
Chris@0 94 */
Chris@0 95 size_t getWindowSize() const { return m_wlen; }
Chris@0 96
Chris@0 97 /**
Chris@16 98 * Get the stretch ratio.
Chris@15 99 */
Chris@16 100 float getRatio() const { return float(m_n2) / float(m_n1); }
Chris@16 101
Chris@16 102 /**
Chris@16 103 * Return whether this time stretcher will attempt to sharpen transients.
Chris@16 104 */
Chris@16 105 bool getSharpening() const { return m_sharpen; }
Chris@15 106
Chris@15 107 /**
Chris@26 108 * Return the number of channels for this time stretcher.
Chris@26 109 */
Chris@26 110 size_t getChannelCount() const { return m_channels; }
Chris@26 111
Chris@26 112 /**
Chris@15 113 * Get the latency added by the time stretcher, in sample frames.
Chris@23 114 * This will be exact if transient sharpening is off, or approximate
Chris@23 115 * if it is on.
Chris@15 116 */
Chris@0 117 size_t getProcessingLatency() const;
Chris@0 118
Chris@0 119 protected:
Chris@13 120 /**
Chris@23 121 * Process a single phase vocoder frame from "in" into
Chris@23 122 * m_freq[channel].
Chris@13 123 */
Chris@23 124 void analyseBlock(size_t channel, float *in); // into m_freq[channel]
Chris@20 125
Chris@23 126 /**
Chris@23 127 * Examine m_freq[0..m_channels-1] and return whether a percussive
Chris@23 128 * transient is found.
Chris@23 129 */
Chris@23 130 bool isTransient();
Chris@20 131
Chris@23 132 /**
Chris@23 133 * Resynthesise from m_freq[channel] adding in to "out",
Chris@23 134 * adjusting phases on the basis of a prior step size of lastStep.
Chris@23 135 * Also add the window shape in to the modulation array (if
Chris@23 136 * present) -- for use in ensuring the output has the correct
Chris@23 137 * magnitude afterwards.
Chris@23 138 */
Chris@20 139 void synthesiseBlock(size_t channel, float *out, float *modulation,
Chris@20 140 size_t lastStep);
Chris@0 141
Chris@25 142 void initialise();
Chris@25 143 void calculateParameters();
Chris@25 144 void cleanup();
Chris@25 145
Chris@32 146 bool shouldSharpen() {
Chris@32 147 return m_sharpen && (m_ratio > 0.25);
Chris@32 148 }
Chris@32 149
Chris@22 150 size_t m_sampleRate;
Chris@16 151 size_t m_channels;
Chris@31 152 size_t m_maxOutputBlockSize;
Chris@12 153 float m_ratio;
Chris@16 154 bool m_sharpen;
Chris@0 155 size_t m_n1;
Chris@0 156 size_t m_n2;
Chris@0 157 size_t m_wlen;
Chris@20 158 Window<float> *m_analysisWindow;
Chris@20 159 Window<float> *m_synthesisWindow;
Chris@0 160
Chris@21 161 int m_totalCount;
Chris@21 162 int m_transientCount;
Chris@21 163 int m_n2sum;
Chris@21 164
Chris@16 165 float **m_prevPhase;
Chris@16 166 float **m_prevAdjustedPhase;
Chris@16 167
Chris@20 168 float *m_prevTransientMag;
Chris@21 169 int m_prevTransientScore;
Chris@22 170 int m_transientThreshold;
Chris@20 171 bool m_prevTransient;
Chris@20 172
Chris@20 173 float *m_tempbuf;
Chris@20 174 float **m_time;
Chris@100 175 fftf_complex **m_freq;
Chris@100 176 fftf_plan *m_plan;
Chris@100 177 fftf_plan *m_iplan;
Chris@0 178
Chris@16 179 RingBuffer<float> **m_inbuf;
Chris@16 180 RingBuffer<float> **m_outbuf;
Chris@16 181 float **m_mashbuf;
Chris@13 182 float *m_modulationbuf;
Chris@25 183
Chris@25 184 QMutex *m_mutex;
Chris@0 185 };
Chris@0 186
Chris@0 187 #endif