annotate audioio/PhaseVocoderTimeStretcher.cpp @ 14:085f34c73939

* IntegerTimeStretcher -> PhaseVocoderTimeStretcher (no longer confined to integer multiples)
author Chris Cannam
date Wed, 13 Sep 2006 11:06:28 +0000
parents audioio/IntegerTimeStretcher.cpp@00ed645f4175
children cc566264c935
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@0 7 This file copyright 2006 Chris Cannam.
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 #include "PhaseVocoderTimeStretcher.h"
Chris@0 17
Chris@0 18 #include <iostream>
Chris@0 19 #include <cassert>
Chris@0 20
Chris@14 21 //#define DEBUG_PHASE_VOCODER_TIME_STRETCHER 1
Chris@0 22
Chris@14 23 PhaseVocoderTimeStretcher::PhaseVocoderTimeStretcher(float ratio,
Chris@0 24 size_t maxProcessInputBlockSize,
Chris@0 25 size_t inputIncrement,
Chris@0 26 size_t windowSize,
Chris@0 27 WindowType windowType) :
Chris@0 28 m_ratio(ratio),
Chris@0 29 m_n1(inputIncrement),
Chris@12 30 m_n2(lrintf(m_n1 * ratio)),
Chris@0 31 m_wlen(std::max(windowSize, m_n2 * 2)),
Chris@0 32 m_inbuf(m_wlen),
Chris@12 33 m_outbuf(maxProcessInputBlockSize * ratio + 1024) //!!!
Chris@0 34 {
Chris@0 35 m_window = new Window<float>(windowType, m_wlen),
Chris@0 36
Chris@0 37 m_time = (fftwf_complex *)fftwf_malloc(sizeof(fftwf_complex) * m_wlen);
Chris@0 38 m_freq = (fftwf_complex *)fftwf_malloc(sizeof(fftwf_complex) * m_wlen);
Chris@0 39 m_dbuf = (float *)fftwf_malloc(sizeof(float) * m_wlen);
Chris@12 40 m_mashbuf = (float *)fftwf_malloc(sizeof(float) * m_wlen);
Chris@13 41 m_modulationbuf = (float *)fftwf_malloc(sizeof(float) * m_wlen);
Chris@12 42 m_prevPhase = (float *)fftwf_malloc(sizeof(float) * m_wlen);
Chris@12 43 m_prevAdjustedPhase = (float *)fftwf_malloc(sizeof(float) * m_wlen);
Chris@0 44
Chris@0 45 m_plan = fftwf_plan_dft_1d(m_wlen, m_time, m_freq, FFTW_FORWARD, FFTW_ESTIMATE);
Chris@0 46 m_iplan = fftwf_plan_dft_c2r_1d(m_wlen, m_freq, m_dbuf, FFTW_ESTIMATE);
Chris@0 47
Chris@0 48 for (int i = 0; i < m_wlen; ++i) {
Chris@0 49 m_mashbuf[i] = 0.0;
Chris@13 50 m_modulationbuf[i] = 0.0;
Chris@12 51 m_prevPhase[i] = 0.0;
Chris@12 52 m_prevAdjustedPhase[i] = 0.0;
Chris@0 53 }
Chris@0 54 }
Chris@0 55
Chris@14 56 PhaseVocoderTimeStretcher::~PhaseVocoderTimeStretcher()
Chris@0 57 {
Chris@14 58 std::cerr << "PhaseVocoderTimeStretcher::~PhaseVocoderTimeStretcher" << std::endl;
Chris@0 59
Chris@0 60 fftwf_destroy_plan(m_plan);
Chris@0 61 fftwf_destroy_plan(m_iplan);
Chris@0 62
Chris@0 63 fftwf_free(m_time);
Chris@0 64 fftwf_free(m_freq);
Chris@0 65 fftwf_free(m_dbuf);
Chris@12 66 fftwf_free(m_mashbuf);
Chris@13 67 fftwf_free(m_modulationbuf);
Chris@12 68 fftwf_free(m_prevPhase);
Chris@12 69 fftwf_free(m_prevAdjustedPhase);
Chris@0 70
Chris@0 71 delete m_window;
Chris@0 72 }
Chris@0 73
Chris@0 74 size_t
Chris@14 75 PhaseVocoderTimeStretcher::getProcessingLatency() const
Chris@0 76 {
Chris@0 77 return getWindowSize() - getInputIncrement();
Chris@0 78 }
Chris@0 79
Chris@0 80 void
Chris@14 81 PhaseVocoderTimeStretcher::process(float *input, float *output, size_t samples)
Chris@0 82 {
Chris@0 83 // We need to add samples from input to our internal buffer. When
Chris@0 84 // we have m_windowSize samples in the buffer, we can process it,
Chris@0 85 // move the samples back by m_n1 and write the output onto our
Chris@0 86 // internal output buffer. If we have (samples * ratio) samples
Chris@0 87 // in that, we can write m_n2 of them back to output and return
Chris@0 88 // (otherwise we have to write zeroes).
Chris@0 89
Chris@0 90 // When we process, we write m_wlen to our fixed output buffer
Chris@0 91 // (m_mashbuf). We then pull out the first m_n2 samples from that
Chris@0 92 // buffer, push them into the output ring buffer, and shift
Chris@0 93 // m_mashbuf left by that amount.
Chris@0 94
Chris@0 95 // The processing latency is then m_wlen - m_n2.
Chris@0 96
Chris@0 97 size_t consumed = 0;
Chris@0 98
Chris@14 99 #ifdef DEBUG_PHASE_VOCODER_TIME_STRETCHER
Chris@14 100 std::cerr << "PhaseVocoderTimeStretcher::process(" << samples << ", consumed = " << consumed << "), writable " << m_inbuf.getWriteSpace() <<", readable "<< m_outbuf.getReadSpace() << std::endl;
Chris@0 101 #endif
Chris@0 102
Chris@0 103 while (consumed < samples) {
Chris@0 104
Chris@0 105 size_t writable = m_inbuf.getWriteSpace();
Chris@0 106 writable = std::min(writable, samples - consumed);
Chris@0 107
Chris@0 108 if (writable == 0) {
Chris@0 109 //!!! then what? I don't think this should happen, but
Chris@14 110 std::cerr << "WARNING: PhaseVocoderTimeStretcher::process: writable == 0" << std::endl;
Chris@0 111 break;
Chris@0 112 }
Chris@0 113
Chris@14 114 #ifdef DEBUG_PHASE_VOCODER_TIME_STRETCHER
Chris@0 115 std::cerr << "writing " << writable << " from index " << consumed << " to inbuf, consumed will be " << consumed + writable << std::endl;
Chris@0 116 #endif
Chris@0 117 m_inbuf.write(input + consumed, writable);
Chris@0 118 consumed += writable;
Chris@0 119
Chris@0 120 while (m_inbuf.getReadSpace() >= m_wlen &&
Chris@0 121 m_outbuf.getWriteSpace() >= m_n2) {
Chris@0 122
Chris@0 123 // We know we have at least m_wlen samples available
Chris@0 124 // in m_inbuf. We need to peek m_wlen of them for
Chris@0 125 // processing, and then read m_n1 to advance the read
Chris@0 126 // pointer.
Chris@0 127
Chris@0 128 size_t got = m_inbuf.peek(m_dbuf, m_wlen);
Chris@0 129 assert(got == m_wlen);
Chris@0 130
Chris@13 131 processBlock(m_dbuf, m_mashbuf, m_modulationbuf);
Chris@0 132
Chris@14 133 #ifdef DEBUG_PHASE_VOCODER_TIME_STRETCHER
Chris@0 134 std::cerr << "writing first " << m_n2 << " from mashbuf, skipping " << m_n1 << " on inbuf " << std::endl;
Chris@0 135 #endif
Chris@0 136 m_inbuf.skip(m_n1);
Chris@13 137
Chris@13 138 for (size_t i = 0; i < m_n2; ++i) {
Chris@13 139 if (m_modulationbuf[i] > 0.f) {
Chris@13 140 m_mashbuf[i] /= m_modulationbuf[i];
Chris@13 141 }
Chris@13 142 }
Chris@13 143
Chris@0 144 m_outbuf.write(m_mashbuf, m_n2);
Chris@0 145
Chris@0 146 for (size_t i = 0; i < m_wlen - m_n2; ++i) {
Chris@0 147 m_mashbuf[i] = m_mashbuf[i + m_n2];
Chris@13 148 m_modulationbuf[i] = m_modulationbuf[i + m_n2];
Chris@0 149 }
Chris@13 150
Chris@0 151 for (size_t i = m_wlen - m_n2; i < m_wlen; ++i) {
Chris@0 152 m_mashbuf[i] = 0.0f;
Chris@13 153 m_modulationbuf[i] = 0.0f;
Chris@0 154 }
Chris@0 155 }
Chris@0 156
Chris@14 157 // std::cerr << "WARNING: PhaseVocoderTimeStretcher::process: writespace not enough for output increment (" << m_outbuf.getWriteSpace() << " < " << m_n2 << ")" << std::endl;
Chris@0 158 // }
Chris@0 159
Chris@14 160 #ifdef DEBUG_PHASE_VOCODER_TIME_STRETCHER
Chris@0 161 std::cerr << "loop ended: inbuf read space " << m_inbuf.getReadSpace() << ", outbuf write space " << m_outbuf.getWriteSpace() << std::endl;
Chris@0 162 #endif
Chris@0 163 }
Chris@0 164
Chris@12 165 size_t toRead = lrintf(samples * m_ratio);
Chris@12 166
Chris@12 167 if (m_outbuf.getReadSpace() < toRead) {
Chris@14 168 std::cerr << "WARNING: PhaseVocoderTimeStretcher::process: not enough data (yet?) (" << m_outbuf.getReadSpace() << " < " << toRead << ")" << std::endl;
Chris@12 169 size_t fill = toRead - m_outbuf.getReadSpace();
Chris@0 170 for (size_t i = 0; i < fill; ++i) {
Chris@0 171 output[i] = 0.0;
Chris@0 172 }
Chris@0 173 m_outbuf.read(output + fill, m_outbuf.getReadSpace());
Chris@0 174 } else {
Chris@14 175 #ifdef DEBUG_PHASE_VOCODER_TIME_STRETCHER
Chris@12 176 std::cerr << "enough data - writing " << toRead << " from outbuf" << std::endl;
Chris@0 177 #endif
Chris@12 178 m_outbuf.read(output, toRead);
Chris@0 179 }
Chris@0 180
Chris@14 181 #ifdef DEBUG_PHASE_VOCODER_TIME_STRETCHER
Chris@14 182 std::cerr << "PhaseVocoderTimeStretcher::process returning" << std::endl;
Chris@0 183 #endif
Chris@0 184 }
Chris@0 185
Chris@0 186 void
Chris@14 187 PhaseVocoderTimeStretcher::processBlock(float *buf, float *out, float *modulation)
Chris@0 188 {
Chris@0 189 size_t i;
Chris@0 190
Chris@0 191 // buf contains m_wlen samples; out contains enough space for
Chris@0 192 // m_wlen * ratio samples (we mix into out, rather than replacing)
Chris@0 193
Chris@14 194 #ifdef DEBUG_PHASE_VOCODER_TIME_STRETCHER
Chris@14 195 std::cerr << "PhaseVocoderTimeStretcher::processBlock" << std::endl;
Chris@0 196 #endif
Chris@0 197
Chris@0 198 m_window->cut(buf);
Chris@0 199
Chris@0 200 for (i = 0; i < m_wlen/2; ++i) {
Chris@0 201 float temp = buf[i];
Chris@0 202 buf[i] = buf[i + m_wlen/2];
Chris@0 203 buf[i + m_wlen/2] = temp;
Chris@0 204 }
Chris@0 205
Chris@0 206 for (i = 0; i < m_wlen; ++i) {
Chris@0 207 m_time[i][0] = buf[i];
Chris@0 208 m_time[i][1] = 0.0;
Chris@0 209 }
Chris@0 210
Chris@0 211 fftwf_execute(m_plan); // m_time -> m_freq
Chris@0 212
Chris@0 213 for (i = 0; i < m_wlen; ++i) {
Chris@0 214
Chris@0 215 float mag = sqrtf(m_freq[i][0] * m_freq[i][0] +
Chris@0 216 m_freq[i][1] * m_freq[i][1]);
Chris@0 217
Chris@12 218 float phase = princargf(atan2f(m_freq[i][1], m_freq[i][0]));
Chris@12 219
Chris@12 220 float omega = (2 * M_PI * m_n1 * i) / m_wlen;
Chris@0 221
Chris@12 222 float expectedPhase = m_prevPhase[i] + omega;
Chris@12 223
Chris@12 224 float phaseError = princargf(phase - expectedPhase);
Chris@12 225
Chris@12 226 float phaseIncrement = (omega + phaseError) / m_n1;
Chris@12 227
Chris@12 228 float adjustedPhase = m_prevAdjustedPhase[i] + m_n2 * phaseIncrement;
Chris@0 229
Chris@12 230 float real = mag * cosf(adjustedPhase);
Chris@12 231 float imag = mag * sinf(adjustedPhase);
Chris@0 232 m_freq[i][0] = real;
Chris@0 233 m_freq[i][1] = imag;
Chris@12 234
Chris@12 235 m_prevPhase[i] = phase;
Chris@12 236 m_prevAdjustedPhase[i] = adjustedPhase;
Chris@0 237 }
Chris@0 238
Chris@0 239 fftwf_execute(m_iplan); // m_freq -> in, inverse fft
Chris@0 240
Chris@0 241 for (i = 0; i < m_wlen/2; ++i) {
Chris@0 242 float temp = buf[i] / m_wlen;
Chris@0 243 buf[i] = buf[i + m_wlen/2] / m_wlen;
Chris@0 244 buf[i + m_wlen/2] = temp;
Chris@0 245 }
Chris@0 246
Chris@0 247 m_window->cut(buf);
Chris@13 248 /*
Chris@0 249 int div = m_wlen / m_n2;
Chris@0 250 if (div > 1) div /= 2;
Chris@0 251 for (i = 0; i < m_wlen; ++i) {
Chris@0 252 buf[i] /= div;
Chris@0 253 }
Chris@13 254 */
Chris@0 255 for (i = 0; i < m_wlen; ++i) {
Chris@0 256 out[i] += buf[i];
Chris@13 257 modulation[i] += m_window->getValue(i);
Chris@0 258 }
Chris@0 259 }