annotate vamp-sdk/hostext/PluginInputDomainAdapter.cpp @ 84:af0d86b8b692

* Fix failure to default correctly if plugin output unspecified
author cannam
date Fri, 28 Sep 2007 09:53:46 +0000
parents 64697dca0d48
children c94c066a4897
rev   line source
cannam@64 1 /* -*- c-basic-offset: 4 indent-tabs-mode: nil -*- vi:set ts=8 sts=4 sw=4: */
cannam@64 2
cannam@64 3 /*
cannam@64 4 Vamp
cannam@64 5
cannam@64 6 An API for audio analysis and feature extraction plugins.
cannam@64 7
cannam@64 8 Centre for Digital Music, Queen Mary, University of London.
cannam@71 9 Copyright 2006-2007 Chris Cannam and QMUL.
cannam@64 10
cannam@64 11 Permission is hereby granted, free of charge, to any person
cannam@64 12 obtaining a copy of this software and associated documentation
cannam@64 13 files (the "Software"), to deal in the Software without
cannam@64 14 restriction, including without limitation the rights to use, copy,
cannam@64 15 modify, merge, publish, distribute, sublicense, and/or sell copies
cannam@64 16 of the Software, and to permit persons to whom the Software is
cannam@64 17 furnished to do so, subject to the following conditions:
cannam@64 18
cannam@64 19 The above copyright notice and this permission notice shall be
cannam@64 20 included in all copies or substantial portions of the Software.
cannam@64 21
cannam@64 22 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
cannam@64 23 EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
cannam@64 24 MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
cannam@64 25 NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR
cannam@64 26 ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF
cannam@64 27 CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
cannam@64 28 WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
cannam@64 29
cannam@64 30 Except as contained in this notice, the names of the Centre for
cannam@64 31 Digital Music; Queen Mary, University of London; and Chris Cannam
cannam@64 32 shall not be used in advertising or otherwise to promote the sale,
cannam@64 33 use or other dealings in this Software without prior written
cannam@64 34 authorization.
cannam@64 35 */
cannam@64 36
cannam@64 37 #include "PluginInputDomainAdapter.h"
cannam@64 38
cannam@64 39 #include <cmath>
cannam@64 40
cannam@64 41 namespace Vamp {
cannam@64 42
cannam@64 43 namespace HostExt {
cannam@64 44
cannam@70 45 class PluginInputDomainAdapter::Impl
cannam@70 46 {
cannam@70 47 public:
cannam@70 48 Impl(Plugin *plugin, float inputSampleRate);
cannam@70 49 ~Impl();
cannam@70 50
cannam@70 51 bool initialise(size_t channels, size_t stepSize, size_t blockSize);
cannam@70 52
cannam@70 53 size_t getPreferredStepSize() const;
cannam@70 54 size_t getPreferredBlockSize() const;
cannam@70 55
cannam@70 56 FeatureSet process(const float *const *inputBuffers, RealTime timestamp);
cannam@70 57
cannam@70 58 protected:
cannam@70 59 Plugin *m_plugin;
cannam@70 60 float m_inputSampleRate;
cannam@70 61 size_t m_channels;
cannam@70 62 size_t m_blockSize;
cannam@70 63 float **m_freqbuf;
cannam@70 64 double *m_ri;
cannam@70 65 double *m_ro;
cannam@70 66 double *m_io;
cannam@70 67
cannam@70 68 void fft(unsigned int n, bool inverse,
cannam@70 69 double *ri, double *ii, double *ro, double *io);
cannam@70 70
cannam@70 71 size_t makeBlockSizeAcceptable(size_t) const;
cannam@70 72 };
cannam@70 73
cannam@64 74 PluginInputDomainAdapter::PluginInputDomainAdapter(Plugin *plugin) :
cannam@70 75 PluginWrapper(plugin)
cannam@70 76 {
cannam@70 77 m_impl = new Impl(plugin, m_inputSampleRate);
cannam@70 78 }
cannam@70 79
cannam@70 80 PluginInputDomainAdapter::~PluginInputDomainAdapter()
cannam@70 81 {
cannam@70 82 delete m_impl;
cannam@70 83 }
cannam@70 84
cannam@70 85 bool
cannam@70 86 PluginInputDomainAdapter::initialise(size_t channels, size_t stepSize, size_t blockSize)
cannam@70 87 {
cannam@70 88 return m_impl->initialise(channels, stepSize, blockSize);
cannam@70 89 }
cannam@70 90
cannam@70 91 Plugin::InputDomain
cannam@70 92 PluginInputDomainAdapter::getInputDomain() const
cannam@70 93 {
cannam@70 94 return TimeDomain;
cannam@70 95 }
cannam@70 96
cannam@70 97 size_t
cannam@70 98 PluginInputDomainAdapter::getPreferredStepSize() const
cannam@70 99 {
cannam@70 100 return m_impl->getPreferredStepSize();
cannam@70 101 }
cannam@70 102
cannam@70 103 size_t
cannam@70 104 PluginInputDomainAdapter::getPreferredBlockSize() const
cannam@70 105 {
cannam@70 106 return m_impl->getPreferredBlockSize();
cannam@70 107 }
cannam@70 108
cannam@70 109 Plugin::FeatureSet
cannam@70 110 PluginInputDomainAdapter::process(const float *const *inputBuffers, RealTime timestamp)
cannam@70 111 {
cannam@70 112 return m_impl->process(inputBuffers, timestamp);
cannam@70 113 }
cannam@70 114
cannam@70 115 PluginInputDomainAdapter::Impl::Impl(Plugin *plugin, float inputSampleRate) :
cannam@70 116 m_plugin(plugin),
cannam@70 117 m_inputSampleRate(inputSampleRate),
cannam@64 118 m_channels(0),
cannam@64 119 m_blockSize(0),
cannam@64 120 m_freqbuf(0)
cannam@64 121 {
cannam@64 122 }
cannam@64 123
cannam@70 124 PluginInputDomainAdapter::Impl::~Impl()
cannam@64 125 {
cannam@70 126 // the adapter will delete the plugin
cannam@70 127
cannam@70 128 if (m_channels > 0) {
cannam@70 129 for (size_t c = 0; c < m_channels; ++c) {
cannam@70 130 delete[] m_freqbuf[c];
cannam@70 131 }
cannam@70 132 delete[] m_freqbuf;
cannam@70 133 delete[] m_ri;
cannam@70 134 delete[] m_ro;
cannam@70 135 delete[] m_io;
cannam@70 136 }
cannam@64 137 }
cannam@64 138
cannam@64 139 bool
cannam@70 140 PluginInputDomainAdapter::Impl::initialise(size_t channels, size_t stepSize, size_t blockSize)
cannam@64 141 {
cannam@64 142 if (m_plugin->getInputDomain() == TimeDomain) {
cannam@64 143
cannam@64 144 m_blockSize = blockSize;
cannam@64 145 m_channels = channels;
cannam@64 146
cannam@64 147 return m_plugin->initialise(channels, stepSize, blockSize);
cannam@64 148 }
cannam@64 149
cannam@64 150 if (blockSize < 2) {
cannam@70 151 std::cerr << "ERROR: Vamp::HostExt::PluginInputDomainAdapter::Impl::initialise: blocksize < 2 not supported" << std::endl;
cannam@64 152 return false;
cannam@64 153 }
cannam@64 154
cannam@64 155 if (blockSize & (blockSize-1)) {
cannam@70 156 std::cerr << "ERROR: Vamp::HostExt::PluginInputDomainAdapter::Impl::initialise: non-power-of-two\nblocksize " << blockSize << " not supported" << std::endl;
cannam@64 157 return false;
cannam@64 158 }
cannam@64 159
cannam@64 160 if (m_channels > 0) {
cannam@64 161 for (size_t c = 0; c < m_channels; ++c) {
cannam@64 162 delete[] m_freqbuf[c];
cannam@64 163 }
cannam@64 164 delete[] m_freqbuf;
cannam@64 165 delete[] m_ri;
cannam@64 166 delete[] m_ro;
cannam@64 167 delete[] m_io;
cannam@64 168 }
cannam@64 169
cannam@64 170 m_blockSize = blockSize;
cannam@64 171 m_channels = channels;
cannam@64 172
cannam@64 173 m_freqbuf = new float *[m_channels];
cannam@64 174 for (size_t c = 0; c < m_channels; ++c) {
cannam@64 175 m_freqbuf[c] = new float[m_blockSize + 2];
cannam@64 176 }
cannam@64 177 m_ri = new double[m_blockSize];
cannam@64 178 m_ro = new double[m_blockSize];
cannam@64 179 m_io = new double[m_blockSize];
cannam@64 180
cannam@64 181 return m_plugin->initialise(channels, stepSize, blockSize);
cannam@64 182 }
cannam@64 183
cannam@64 184 size_t
cannam@70 185 PluginInputDomainAdapter::Impl::getPreferredStepSize() const
cannam@64 186 {
cannam@64 187 size_t step = m_plugin->getPreferredStepSize();
cannam@64 188
cannam@64 189 if (step == 0 && (m_plugin->getInputDomain() == FrequencyDomain)) {
cannam@64 190 step = getPreferredBlockSize() / 2;
cannam@64 191 }
cannam@64 192
cannam@64 193 return step;
cannam@64 194 }
cannam@64 195
cannam@64 196 size_t
cannam@70 197 PluginInputDomainAdapter::Impl::getPreferredBlockSize() const
cannam@64 198 {
cannam@64 199 size_t block = m_plugin->getPreferredBlockSize();
cannam@64 200
cannam@64 201 if (m_plugin->getInputDomain() == FrequencyDomain) {
cannam@64 202 if (block == 0) {
cannam@64 203 block = 1024;
cannam@64 204 } else {
cannam@64 205 block = makeBlockSizeAcceptable(block);
cannam@64 206 }
cannam@64 207 }
cannam@64 208
cannam@64 209 return block;
cannam@64 210 }
cannam@64 211
cannam@64 212 size_t
cannam@70 213 PluginInputDomainAdapter::Impl::makeBlockSizeAcceptable(size_t blockSize) const
cannam@64 214 {
cannam@64 215 if (blockSize < 2) {
cannam@64 216
cannam@70 217 std::cerr << "WARNING: Vamp::HostExt::PluginInputDomainAdapter::Impl::initialise: blocksize < 2 not" << std::endl
cannam@64 218 << "supported, increasing from " << blockSize << " to 2" << std::endl;
cannam@64 219 blockSize = 2;
cannam@64 220
cannam@64 221 } else if (blockSize & (blockSize-1)) {
cannam@64 222
cannam@64 223 // not a power of two, can't handle that with our current fft
cannam@64 224 // implementation
cannam@64 225
cannam@64 226 size_t nearest = blockSize;
cannam@64 227 size_t power = 0;
cannam@64 228 while (nearest > 1) {
cannam@64 229 nearest >>= 1;
cannam@64 230 ++power;
cannam@64 231 }
cannam@64 232 nearest = 1;
cannam@64 233 while (power) {
cannam@64 234 nearest <<= 1;
cannam@64 235 --power;
cannam@64 236 }
cannam@64 237
cannam@64 238 if (blockSize - nearest > (nearest*2) - blockSize) {
cannam@64 239 nearest = nearest*2;
cannam@64 240 }
cannam@64 241
cannam@70 242 std::cerr << "WARNING: Vamp::HostExt::PluginInputDomainAdapter::Impl::initialise: non-power-of-two\nblocksize " << blockSize << " not supported, using blocksize " << nearest << " instead" << std::endl;
cannam@64 243 blockSize = nearest;
cannam@64 244 }
cannam@64 245
cannam@64 246 return blockSize;
cannam@64 247 }
cannam@64 248
cannam@68 249 // for some visual studii apparently
cannam@68 250 #ifndef M_PI
cannam@68 251 #define M_PI 3.14159265358979232846
cannam@68 252 #endif
cannam@68 253
cannam@64 254 Plugin::FeatureSet
cannam@70 255 PluginInputDomainAdapter::Impl::process(const float *const *inputBuffers,
cannam@70 256 RealTime timestamp)
cannam@64 257 {
cannam@64 258 if (m_plugin->getInputDomain() == TimeDomain) {
cannam@64 259 return m_plugin->process(inputBuffers, timestamp);
cannam@64 260 }
cannam@64 261
cannam@64 262 // The timestamp supplied should be (according to the Vamp::Plugin
cannam@64 263 // spec) the time of the start of the time-domain input block.
cannam@64 264 // However, we want to pass to the plugin an FFT output calculated
cannam@64 265 // from the block of samples _centred_ on that timestamp.
cannam@64 266 //
cannam@64 267 // We have two options:
cannam@64 268 //
cannam@64 269 // 1. Buffer the input, calculating the fft of the values at the
cannam@64 270 // passed-in block minus blockSize/2 rather than starting at the
cannam@64 271 // passed-in block. So each time we call process on the plugin,
cannam@64 272 // we are passing in the same timestamp as was passed to our own
cannam@64 273 // process plugin, but not (the frequency domain representation
cannam@64 274 // of) the same set of samples. Advantages: avoids confusion in
cannam@64 275 // the host by ensuring the returned values have timestamps
cannam@64 276 // comparable with that passed in to this function (in fact this
cannam@64 277 // is pretty much essential for one-value-per-block outputs);
cannam@64 278 // consistent with hosts such as SV that deal with the
cannam@64 279 // frequency-domain transform themselves. Disadvantages: means
cannam@64 280 // making the not necessarily correct assumption that the samples
cannam@64 281 // preceding the first official block are all zero (or some other
cannam@64 282 // known value).
cannam@64 283 //
cannam@64 284 // 2. Increase the passed-in timestamps by half the blocksize. So
cannam@64 285 // when we call process, we are passing in the frequency domain
cannam@64 286 // representation of the same set of samples as passed to us, but
cannam@64 287 // with a different timestamp. Advantages: simplicity; avoids
cannam@64 288 // iffy assumption mentioned above. Disadvantages: inconsistency
cannam@64 289 // with SV in cases where stepSize != blockSize/2; potential
cannam@64 290 // confusion arising from returned timestamps being calculated
cannam@64 291 // from the adjusted input timestamps rather than the original
cannam@64 292 // ones (and inaccuracy where the returned timestamp is implied,
cannam@64 293 // as in one-value-per-block).
cannam@64 294 //
cannam@64 295 // Neither way is ideal, but I don't think either is strictly
cannam@64 296 // incorrect either. I think this is just a case where the same
cannam@64 297 // plugin can legitimately produce differing results from the same
cannam@64 298 // input data, depending on how that data is packaged.
cannam@64 299 //
cannam@64 300 // We'll go for option 2, adjusting the timestamps. Note in
cannam@64 301 // particular that this means some results can differ from those
cannam@64 302 // produced by SV.
cannam@64 303
cannam@65 304 // std::cerr << "PluginInputDomainAdapter: sampleRate " << m_inputSampleRate << ", blocksize " << m_blockSize << ", adjusting time from " << timestamp;
cannam@64 305
cannam@68 306 timestamp = timestamp + RealTime::frame2RealTime
cannam@68 307 (m_blockSize/2, int(m_inputSampleRate + 0.5));
cannam@64 308
cannam@65 309 // std::cerr << " to " << timestamp << std::endl;
cannam@64 310
cannam@64 311 for (size_t c = 0; c < m_channels; ++c) {
cannam@64 312
cannam@64 313 for (size_t i = 0; i < m_blockSize; ++i) {
cannam@64 314 // Hanning window
cannam@64 315 m_ri[i] = double(inputBuffers[c][i])
cannam@64 316 * (0.50 - 0.50 * cos((2 * M_PI * i)
cannam@64 317 / m_blockSize));
cannam@64 318 }
cannam@64 319
cannam@64 320 for (size_t i = 0; i < m_blockSize/2; ++i) {
cannam@64 321 // FFT shift
cannam@64 322 double value = m_ri[i];
cannam@64 323 m_ri[i] = m_ri[i + m_blockSize/2];
cannam@64 324 m_ri[i + m_blockSize/2] = value;
cannam@64 325 }
cannam@64 326
cannam@64 327 fft(m_blockSize, false, m_ri, 0, m_ro, m_io);
cannam@64 328
cannam@64 329 for (size_t i = 0; i <= m_blockSize/2; ++i) {
cannam@64 330 m_freqbuf[c][i * 2] = m_ro[i];
cannam@64 331 m_freqbuf[c][i * 2 + 1] = m_io[i];
cannam@64 332 }
cannam@64 333 }
cannam@64 334
cannam@64 335 return m_plugin->process(m_freqbuf, timestamp);
cannam@64 336 }
cannam@64 337
cannam@64 338 void
cannam@70 339 PluginInputDomainAdapter::Impl::fft(unsigned int n, bool inverse,
cannam@70 340 double *ri, double *ii, double *ro, double *io)
cannam@64 341 {
cannam@64 342 if (!ri || !ro || !io) return;
cannam@64 343
cannam@64 344 unsigned int bits;
cannam@64 345 unsigned int i, j, k, m;
cannam@64 346 unsigned int blockSize, blockEnd;
cannam@64 347
cannam@64 348 double tr, ti;
cannam@64 349
cannam@64 350 if (n < 2) return;
cannam@64 351 if (n & (n-1)) return;
cannam@64 352
cannam@64 353 double angle = 2.0 * M_PI;
cannam@64 354 if (inverse) angle = -angle;
cannam@64 355
cannam@64 356 for (i = 0; ; ++i) {
cannam@64 357 if (n & (1 << i)) {
cannam@64 358 bits = i;
cannam@64 359 break;
cannam@64 360 }
cannam@64 361 }
cannam@64 362
cannam@64 363 static unsigned int tableSize = 0;
cannam@64 364 static int *table = 0;
cannam@64 365
cannam@64 366 if (tableSize != n) {
cannam@64 367
cannam@64 368 delete[] table;
cannam@64 369
cannam@64 370 table = new int[n];
cannam@64 371
cannam@64 372 for (i = 0; i < n; ++i) {
cannam@64 373
cannam@64 374 m = i;
cannam@64 375
cannam@64 376 for (j = k = 0; j < bits; ++j) {
cannam@64 377 k = (k << 1) | (m & 1);
cannam@64 378 m >>= 1;
cannam@64 379 }
cannam@64 380
cannam@64 381 table[i] = k;
cannam@64 382 }
cannam@64 383
cannam@64 384 tableSize = n;
cannam@64 385 }
cannam@64 386
cannam@64 387 if (ii) {
cannam@64 388 for (i = 0; i < n; ++i) {
cannam@64 389 ro[table[i]] = ri[i];
cannam@64 390 io[table[i]] = ii[i];
cannam@64 391 }
cannam@64 392 } else {
cannam@64 393 for (i = 0; i < n; ++i) {
cannam@64 394 ro[table[i]] = ri[i];
cannam@64 395 io[table[i]] = 0.0;
cannam@64 396 }
cannam@64 397 }
cannam@64 398
cannam@64 399 blockEnd = 1;
cannam@64 400
cannam@64 401 for (blockSize = 2; blockSize <= n; blockSize <<= 1) {
cannam@64 402
cannam@64 403 double delta = angle / (double)blockSize;
cannam@64 404 double sm2 = -sin(-2 * delta);
cannam@64 405 double sm1 = -sin(-delta);
cannam@64 406 double cm2 = cos(-2 * delta);
cannam@64 407 double cm1 = cos(-delta);
cannam@64 408 double w = 2 * cm1;
cannam@64 409 double ar[3], ai[3];
cannam@64 410
cannam@64 411 for (i = 0; i < n; i += blockSize) {
cannam@64 412
cannam@64 413 ar[2] = cm2;
cannam@64 414 ar[1] = cm1;
cannam@64 415
cannam@64 416 ai[2] = sm2;
cannam@64 417 ai[1] = sm1;
cannam@64 418
cannam@64 419 for (j = i, m = 0; m < blockEnd; j++, m++) {
cannam@64 420
cannam@64 421 ar[0] = w * ar[1] - ar[2];
cannam@64 422 ar[2] = ar[1];
cannam@64 423 ar[1] = ar[0];
cannam@64 424
cannam@64 425 ai[0] = w * ai[1] - ai[2];
cannam@64 426 ai[2] = ai[1];
cannam@64 427 ai[1] = ai[0];
cannam@64 428
cannam@64 429 k = j + blockEnd;
cannam@64 430 tr = ar[0] * ro[k] - ai[0] * io[k];
cannam@64 431 ti = ar[0] * io[k] + ai[0] * ro[k];
cannam@64 432
cannam@64 433 ro[k] = ro[j] - tr;
cannam@64 434 io[k] = io[j] - ti;
cannam@64 435
cannam@64 436 ro[j] += tr;
cannam@64 437 io[j] += ti;
cannam@64 438 }
cannam@64 439 }
cannam@64 440
cannam@64 441 blockEnd = blockSize;
cannam@64 442 }
cannam@64 443
cannam@64 444 if (inverse) {
cannam@64 445
cannam@64 446 double denom = (double)n;
cannam@64 447
cannam@64 448 for (i = 0; i < n; i++) {
cannam@64 449 ro[i] /= denom;
cannam@64 450 io[i] /= denom;
cannam@64 451 }
cannam@64 452 }
cannam@64 453 }
cannam@64 454
cannam@64 455 }
cannam@64 456
cannam@64 457 }
cannam@64 458