annotate transform/FeatureExtractionPluginTransform.cpp @ 0:cd5d7ff8ef38

* Reorganising code base. This revision will not compile.
author Chris Cannam
date Mon, 31 Jul 2006 12:03:45 +0000
parents
children 40116f709d3b
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@0 16 #include "FeatureExtractionPluginTransform.h"
Chris@0 17
Chris@0 18 #include "plugin/FeatureExtractionPluginFactory.h"
Chris@0 19 #include "plugin/PluginXml.h"
Chris@0 20 #include "vamp-sdk/Plugin.h"
Chris@0 21
Chris@0 22 #include "base/Model.h"
Chris@0 23 #include "base/Window.h"
Chris@0 24 #include "model/SparseOneDimensionalModel.h"
Chris@0 25 #include "model/SparseTimeValueModel.h"
Chris@0 26 #include "model/DenseThreeDimensionalModel.h"
Chris@0 27 #include "model/DenseTimeValueModel.h"
Chris@0 28 #include "model/NoteModel.h"
Chris@0 29 #include "fileio/FFTFuzzyAdapter.h"
Chris@0 30
Chris@0 31 #include <fftw3.h>
Chris@0 32
Chris@0 33 #include <iostream>
Chris@0 34
Chris@0 35 FeatureExtractionPluginTransform::FeatureExtractionPluginTransform(Model *inputModel,
Chris@0 36 QString pluginId,
Chris@0 37 int channel,
Chris@0 38 QString configurationXml,
Chris@0 39 QString outputName) :
Chris@0 40 Transform(inputModel),
Chris@0 41 m_plugin(0),
Chris@0 42 m_channel(channel),
Chris@0 43 m_stepSize(0),
Chris@0 44 m_blockSize(0),
Chris@0 45 m_descriptor(0),
Chris@0 46 m_outputFeatureNo(0)
Chris@0 47 {
Chris@0 48 // std::cerr << "FeatureExtractionPluginTransform::FeatureExtractionPluginTransform: plugin " << pluginId.toStdString() << ", outputName " << outputName.toStdString() << std::endl;
Chris@0 49
Chris@0 50 FeatureExtractionPluginFactory *factory =
Chris@0 51 FeatureExtractionPluginFactory::instanceFor(pluginId);
Chris@0 52
Chris@0 53 if (!factory) {
Chris@0 54 std::cerr << "FeatureExtractionPluginTransform: No factory available for plugin id \""
Chris@0 55 << pluginId.toStdString() << "\"" << std::endl;
Chris@0 56 return;
Chris@0 57 }
Chris@0 58
Chris@0 59 m_plugin = factory->instantiatePlugin(pluginId, m_input->getSampleRate());
Chris@0 60
Chris@0 61 if (!m_plugin) {
Chris@0 62 std::cerr << "FeatureExtractionPluginTransform: Failed to instantiate plugin \""
Chris@0 63 << pluginId.toStdString() << "\"" << std::endl;
Chris@0 64 return;
Chris@0 65 }
Chris@0 66
Chris@0 67 if (configurationXml != "") {
Chris@0 68 PluginXml(m_plugin).setParametersFromXml(configurationXml);
Chris@0 69 }
Chris@0 70
Chris@0 71 m_blockSize = m_plugin->getPreferredBlockSize();
Chris@0 72 m_stepSize = m_plugin->getPreferredStepSize();
Chris@0 73
Chris@0 74 if (m_blockSize == 0) m_blockSize = 1024; //!!! todo: ask user
Chris@0 75 if (m_stepSize == 0) m_stepSize = m_blockSize; //!!! likewise
Chris@0 76
Chris@0 77 DenseTimeValueModel *input = getInput();
Chris@0 78 if (!input) return;
Chris@0 79
Chris@0 80 size_t channelCount = input->getChannelCount();
Chris@0 81 if (m_plugin->getMaxChannelCount() < channelCount) {
Chris@0 82 channelCount = 1;
Chris@0 83 }
Chris@0 84 if (m_plugin->getMinChannelCount() > channelCount) {
Chris@0 85 std::cerr << "FeatureExtractionPluginTransform:: "
Chris@0 86 << "Can't provide enough channels to plugin (plugin min "
Chris@0 87 << m_plugin->getMinChannelCount() << ", max "
Chris@0 88 << m_plugin->getMaxChannelCount() << ", input model has "
Chris@0 89 << input->getChannelCount() << ")" << std::endl;
Chris@0 90 return;
Chris@0 91 }
Chris@0 92
Chris@0 93 if (!m_plugin->initialise(channelCount, m_stepSize, m_blockSize)) {
Chris@0 94 std::cerr << "FeatureExtractionPluginTransform: Plugin "
Chris@0 95 << m_plugin->getName() << " failed to initialise!" << std::endl;
Chris@0 96 return;
Chris@0 97 }
Chris@0 98
Chris@0 99 Vamp::Plugin::OutputList outputs = m_plugin->getOutputDescriptors();
Chris@0 100
Chris@0 101 if (outputs.empty()) {
Chris@0 102 std::cerr << "FeatureExtractionPluginTransform: Plugin \""
Chris@0 103 << pluginId.toStdString() << "\" has no outputs" << std::endl;
Chris@0 104 return;
Chris@0 105 }
Chris@0 106
Chris@0 107 for (size_t i = 0; i < outputs.size(); ++i) {
Chris@0 108 if (outputName == "" || outputs[i].name == outputName.toStdString()) {
Chris@0 109 m_outputFeatureNo = i;
Chris@0 110 m_descriptor = new Vamp::Plugin::OutputDescriptor
Chris@0 111 (outputs[i]);
Chris@0 112 break;
Chris@0 113 }
Chris@0 114 }
Chris@0 115
Chris@0 116 if (!m_descriptor) {
Chris@0 117 std::cerr << "FeatureExtractionPluginTransform: Plugin \""
Chris@0 118 << pluginId.toStdString() << "\" has no output named \""
Chris@0 119 << outputName.toStdString() << "\"" << std::endl;
Chris@0 120 return;
Chris@0 121 }
Chris@0 122
Chris@0 123 // std::cerr << "FeatureExtractionPluginTransform: output sample type "
Chris@0 124 // << m_descriptor->sampleType << std::endl;
Chris@0 125
Chris@0 126 int binCount = 1;
Chris@0 127 float minValue = 0.0, maxValue = 0.0;
Chris@0 128
Chris@0 129 if (m_descriptor->hasFixedBinCount) {
Chris@0 130 binCount = m_descriptor->binCount;
Chris@0 131 }
Chris@0 132
Chris@0 133 // std::cerr << "FeatureExtractionPluginTransform: output bin count "
Chris@0 134 // << binCount << std::endl;
Chris@0 135
Chris@0 136 if (binCount > 0 && m_descriptor->hasKnownExtents) {
Chris@0 137 minValue = m_descriptor->minValue;
Chris@0 138 maxValue = m_descriptor->maxValue;
Chris@0 139 }
Chris@0 140
Chris@0 141 size_t modelRate = m_input->getSampleRate();
Chris@0 142 size_t modelResolution = 1;
Chris@0 143
Chris@0 144 switch (m_descriptor->sampleType) {
Chris@0 145
Chris@0 146 case Vamp::Plugin::OutputDescriptor::VariableSampleRate:
Chris@0 147 if (m_descriptor->sampleRate != 0.0) {
Chris@0 148 modelResolution = size_t(modelRate / m_descriptor->sampleRate + 0.001);
Chris@0 149 }
Chris@0 150 break;
Chris@0 151
Chris@0 152 case Vamp::Plugin::OutputDescriptor::OneSamplePerStep:
Chris@0 153 modelResolution = m_stepSize;
Chris@0 154 break;
Chris@0 155
Chris@0 156 case Vamp::Plugin::OutputDescriptor::FixedSampleRate:
Chris@0 157 modelRate = size_t(m_descriptor->sampleRate + 0.001);
Chris@0 158 break;
Chris@0 159 }
Chris@0 160
Chris@0 161 if (binCount == 0) {
Chris@0 162
Chris@0 163 m_output = new SparseOneDimensionalModel(modelRate, modelResolution,
Chris@0 164 false);
Chris@0 165
Chris@0 166 } else if (binCount == 1) {
Chris@0 167
Chris@0 168 SparseTimeValueModel *model = new SparseTimeValueModel
Chris@0 169 (modelRate, modelResolution, minValue, maxValue, false);
Chris@0 170 model->setScaleUnits(outputs[m_outputFeatureNo].unit.c_str());
Chris@0 171
Chris@0 172 m_output = model;
Chris@0 173
Chris@0 174 } else if (m_descriptor->sampleType ==
Chris@0 175 Vamp::Plugin::OutputDescriptor::VariableSampleRate) {
Chris@0 176
Chris@0 177 // We don't have a sparse 3D model, so interpret this as a
Chris@0 178 // note model. There's nothing to define which values to use
Chris@0 179 // as which parameters of the note -- for the moment let's
Chris@0 180 // treat the first as pitch, second as duration in frames,
Chris@0 181 // third (if present) as velocity. (Our note model doesn't
Chris@0 182 // yet store velocity.)
Chris@0 183 //!!! todo: ask the user!
Chris@0 184
Chris@0 185 NoteModel *model = new NoteModel
Chris@0 186 (modelRate, modelResolution, minValue, maxValue, false);
Chris@0 187 model->setScaleUnits(outputs[m_outputFeatureNo].unit.c_str());
Chris@0 188
Chris@0 189 m_output = model;
Chris@0 190
Chris@0 191 } else {
Chris@0 192
Chris@0 193 m_output = new DenseThreeDimensionalModel(modelRate, modelResolution,
Chris@0 194 binCount, false);
Chris@0 195
Chris@0 196 if (!m_descriptor->binNames.empty()) {
Chris@0 197 std::vector<QString> names;
Chris@0 198 for (size_t i = 0; i < m_descriptor->binNames.size(); ++i) {
Chris@0 199 names.push_back(m_descriptor->binNames[i].c_str());
Chris@0 200 }
Chris@0 201 (dynamic_cast<DenseThreeDimensionalModel *>(m_output))
Chris@0 202 ->setBinNames(names);
Chris@0 203 }
Chris@0 204 }
Chris@0 205 }
Chris@0 206
Chris@0 207 FeatureExtractionPluginTransform::~FeatureExtractionPluginTransform()
Chris@0 208 {
Chris@0 209 delete m_plugin;
Chris@0 210 delete m_descriptor;
Chris@0 211 }
Chris@0 212
Chris@0 213 DenseTimeValueModel *
Chris@0 214 FeatureExtractionPluginTransform::getInput()
Chris@0 215 {
Chris@0 216 DenseTimeValueModel *dtvm =
Chris@0 217 dynamic_cast<DenseTimeValueModel *>(getInputModel());
Chris@0 218 if (!dtvm) {
Chris@0 219 std::cerr << "FeatureExtractionPluginTransform::getInput: WARNING: Input model is not conformable to DenseTimeValueModel" << std::endl;
Chris@0 220 }
Chris@0 221 return dtvm;
Chris@0 222 }
Chris@0 223
Chris@0 224 void
Chris@0 225 FeatureExtractionPluginTransform::run()
Chris@0 226 {
Chris@0 227 DenseTimeValueModel *input = getInput();
Chris@0 228 if (!input) return;
Chris@0 229
Chris@0 230 if (!m_output) return;
Chris@0 231
Chris@0 232 size_t sampleRate = m_input->getSampleRate();
Chris@0 233
Chris@0 234 size_t channelCount = input->getChannelCount();
Chris@0 235 if (m_plugin->getMaxChannelCount() < channelCount) {
Chris@0 236 channelCount = 1;
Chris@0 237 }
Chris@0 238
Chris@0 239 float **buffers = new float*[channelCount];
Chris@0 240 for (size_t ch = 0; ch < channelCount; ++ch) {
Chris@0 241 buffers[ch] = new float[m_blockSize];
Chris@0 242 }
Chris@0 243
Chris@0 244 bool frequencyDomain = (m_plugin->getInputDomain() ==
Chris@0 245 Vamp::Plugin::FrequencyDomain);
Chris@0 246 std::vector<FFTFuzzyAdapter *> fftAdapters;
Chris@0 247
Chris@0 248 if (frequencyDomain) {
Chris@0 249 for (size_t ch = 0; ch < channelCount; ++ch) {
Chris@0 250 fftAdapters.push_back(new FFTFuzzyAdapter
Chris@0 251 (getInput(),
Chris@0 252 channelCount == 1 ? m_channel : ch,
Chris@0 253 HanningWindow,
Chris@0 254 m_blockSize,
Chris@0 255 m_stepSize,
Chris@0 256 m_blockSize,
Chris@0 257 false));
Chris@0 258 }
Chris@0 259 }
Chris@0 260
Chris@0 261 long startFrame = m_input->getStartFrame();
Chris@0 262 long endFrame = m_input->getEndFrame();
Chris@0 263 long blockFrame = startFrame;
Chris@0 264
Chris@0 265 long prevCompletion = 0;
Chris@0 266
Chris@0 267 while (1) {
Chris@0 268
Chris@0 269 if (frequencyDomain) {
Chris@0 270 if (blockFrame - int(m_blockSize)/2 > endFrame) break;
Chris@0 271 } else {
Chris@0 272 if (blockFrame >= endFrame) break;
Chris@0 273 }
Chris@0 274
Chris@0 275 // std::cerr << "FeatureExtractionPluginTransform::run: blockFrame "
Chris@0 276 // << blockFrame << std::endl;
Chris@0 277
Chris@0 278 long completion =
Chris@0 279 (((blockFrame - startFrame) / m_stepSize) * 99) /
Chris@0 280 ( (endFrame - startFrame) / m_stepSize);
Chris@0 281
Chris@0 282 // channelCount is either m_input->channelCount or 1
Chris@0 283
Chris@0 284 for (size_t ch = 0; ch < channelCount; ++ch) {
Chris@0 285 if (frequencyDomain) {
Chris@0 286 int column = (blockFrame - startFrame) / m_stepSize;
Chris@0 287 for (size_t i = 0; i < m_blockSize/2; ++i) {
Chris@0 288 fftAdapters[ch]->getValuesAt
Chris@0 289 (column, i, buffers[ch][i*2], buffers[ch][i*2+1]);
Chris@0 290 }
Chris@0 291 /*!!!
Chris@0 292 float sum = 0.0;
Chris@0 293 for (size_t i = 0; i < m_blockSize/2; ++i) {
Chris@0 294 sum += buffers[ch][i*2];
Chris@0 295 }
Chris@0 296 if (fabs(sum) < 0.0001) {
Chris@0 297 std::cerr << "WARNING: small sum for column " << column << " (sum is " << sum << ")" << std::endl;
Chris@0 298 }
Chris@0 299 */
Chris@0 300 } else {
Chris@0 301 getFrames(ch, channelCount,
Chris@0 302 blockFrame, m_blockSize, buffers[ch]);
Chris@0 303 }
Chris@0 304 }
Chris@0 305
Chris@0 306 Vamp::Plugin::FeatureSet features = m_plugin->process
Chris@0 307 (buffers, Vamp::RealTime::frame2RealTime(blockFrame, sampleRate));
Chris@0 308
Chris@0 309 for (size_t fi = 0; fi < features[m_outputFeatureNo].size(); ++fi) {
Chris@0 310 Vamp::Plugin::Feature feature =
Chris@0 311 features[m_outputFeatureNo][fi];
Chris@0 312 addFeature(blockFrame, feature);
Chris@0 313 }
Chris@0 314
Chris@0 315 if (blockFrame == startFrame || completion > prevCompletion) {
Chris@0 316 setCompletion(completion);
Chris@0 317 prevCompletion = completion;
Chris@0 318 }
Chris@0 319
Chris@0 320 blockFrame += m_stepSize;
Chris@0 321 }
Chris@0 322
Chris@0 323 Vamp::Plugin::FeatureSet features = m_plugin->getRemainingFeatures();
Chris@0 324
Chris@0 325 for (size_t fi = 0; fi < features[m_outputFeatureNo].size(); ++fi) {
Chris@0 326 Vamp::Plugin::Feature feature =
Chris@0 327 features[m_outputFeatureNo][fi];
Chris@0 328 addFeature(blockFrame, feature);
Chris@0 329 }
Chris@0 330
Chris@0 331 if (frequencyDomain) {
Chris@0 332 for (size_t ch = 0; ch < channelCount; ++ch) {
Chris@0 333 delete fftAdapters[ch];
Chris@0 334 }
Chris@0 335 }
Chris@0 336
Chris@0 337 setCompletion(100);
Chris@0 338 }
Chris@0 339
Chris@0 340 void
Chris@0 341 FeatureExtractionPluginTransform::getFrames(int channel, int channelCount,
Chris@0 342 long startFrame, long size,
Chris@0 343 float *buffer)
Chris@0 344 {
Chris@0 345 long offset = 0;
Chris@0 346
Chris@0 347 if (startFrame < 0) {
Chris@0 348 for (int i = 0; i < size && startFrame + i < 0; ++i) {
Chris@0 349 buffer[i] = 0.0f;
Chris@0 350 }
Chris@0 351 offset = -startFrame;
Chris@0 352 size -= offset;
Chris@0 353 if (size <= 0) return;
Chris@0 354 startFrame = 0;
Chris@0 355 }
Chris@0 356
Chris@0 357 long got = getInput()->getValues
Chris@0 358 ((channelCount == 1 ? m_channel : channel),
Chris@0 359 startFrame, startFrame + size, buffer + offset);
Chris@0 360
Chris@0 361 while (got < size) {
Chris@0 362 buffer[offset + got] = 0.0;
Chris@0 363 ++got;
Chris@0 364 }
Chris@0 365
Chris@0 366 if (m_channel == -1 && channelCount == 1 &&
Chris@0 367 getInput()->getChannelCount() > 1) {
Chris@0 368 // use mean instead of sum, as plugin input
Chris@0 369 int cc = getInput()->getChannelCount();
Chris@0 370 for (long i = 0; i < size; ++i) {
Chris@0 371 buffer[i] /= cc;
Chris@0 372 }
Chris@0 373 }
Chris@0 374 }
Chris@0 375
Chris@0 376 void
Chris@0 377 FeatureExtractionPluginTransform::addFeature(size_t blockFrame,
Chris@0 378 const Vamp::Plugin::Feature &feature)
Chris@0 379 {
Chris@0 380 size_t inputRate = m_input->getSampleRate();
Chris@0 381
Chris@0 382 // std::cerr << "FeatureExtractionPluginTransform::addFeature("
Chris@0 383 // << blockFrame << ")" << std::endl;
Chris@0 384
Chris@0 385 int binCount = 1;
Chris@0 386 if (m_descriptor->hasFixedBinCount) {
Chris@0 387 binCount = m_descriptor->binCount;
Chris@0 388 }
Chris@0 389
Chris@0 390 size_t frame = blockFrame;
Chris@0 391
Chris@0 392 if (m_descriptor->sampleType ==
Chris@0 393 Vamp::Plugin::OutputDescriptor::VariableSampleRate) {
Chris@0 394
Chris@0 395 if (!feature.hasTimestamp) {
Chris@0 396 std::cerr
Chris@0 397 << "WARNING: FeatureExtractionPluginTransform::addFeature: "
Chris@0 398 << "Feature has variable sample rate but no timestamp!"
Chris@0 399 << std::endl;
Chris@0 400 return;
Chris@0 401 } else {
Chris@0 402 frame = Vamp::RealTime::realTime2Frame(feature.timestamp, inputRate);
Chris@0 403 }
Chris@0 404
Chris@0 405 } else if (m_descriptor->sampleType ==
Chris@0 406 Vamp::Plugin::OutputDescriptor::FixedSampleRate) {
Chris@0 407
Chris@0 408 if (feature.hasTimestamp) {
Chris@0 409 //!!! warning: sampleRate may be non-integral
Chris@0 410 frame = Vamp::RealTime::realTime2Frame(feature.timestamp,
Chris@0 411 m_descriptor->sampleRate);
Chris@0 412 } else {
Chris@0 413 frame = m_output->getEndFrame() + 1;
Chris@0 414 }
Chris@0 415 }
Chris@0 416
Chris@0 417 if (binCount == 0) {
Chris@0 418
Chris@0 419 SparseOneDimensionalModel *model = getOutput<SparseOneDimensionalModel>();
Chris@0 420 if (!model) return;
Chris@0 421 model->addPoint(SparseOneDimensionalModel::Point(frame, feature.label.c_str()));
Chris@0 422
Chris@0 423 } else if (binCount == 1) {
Chris@0 424
Chris@0 425 float value = 0.0;
Chris@0 426 if (feature.values.size() > 0) value = feature.values[0];
Chris@0 427
Chris@0 428 SparseTimeValueModel *model = getOutput<SparseTimeValueModel>();
Chris@0 429 if (!model) return;
Chris@0 430 model->addPoint(SparseTimeValueModel::Point(frame, value, feature.label.c_str()));
Chris@0 431
Chris@0 432 } else if (m_descriptor->sampleType ==
Chris@0 433 Vamp::Plugin::OutputDescriptor::VariableSampleRate) {
Chris@0 434
Chris@0 435 float pitch = 0.0;
Chris@0 436 if (feature.values.size() > 0) pitch = feature.values[0];
Chris@0 437
Chris@0 438 float duration = 1;
Chris@0 439 if (feature.values.size() > 1) duration = feature.values[1];
Chris@0 440
Chris@0 441 float velocity = 100;
Chris@0 442 if (feature.values.size() > 2) velocity = feature.values[2];
Chris@0 443
Chris@0 444 NoteModel *model = getOutput<NoteModel>();
Chris@0 445 if (!model) return;
Chris@0 446
Chris@0 447 model->addPoint(NoteModel::Point(frame, pitch, duration, feature.label.c_str()));
Chris@0 448
Chris@0 449 } else {
Chris@0 450
Chris@0 451 DenseThreeDimensionalModel::BinValueSet values = feature.values;
Chris@0 452
Chris@0 453 DenseThreeDimensionalModel *model = getOutput<DenseThreeDimensionalModel>();
Chris@0 454 if (!model) return;
Chris@0 455
Chris@0 456 model->setBinValues(frame, values);
Chris@0 457 }
Chris@0 458 }
Chris@0 459
Chris@0 460 void
Chris@0 461 FeatureExtractionPluginTransform::setCompletion(int completion)
Chris@0 462 {
Chris@0 463 int binCount = 1;
Chris@0 464 if (m_descriptor->hasFixedBinCount) {
Chris@0 465 binCount = m_descriptor->binCount;
Chris@0 466 }
Chris@0 467
Chris@0 468 if (binCount == 0) {
Chris@0 469
Chris@0 470 SparseOneDimensionalModel *model = getOutput<SparseOneDimensionalModel>();
Chris@0 471 if (!model) return;
Chris@0 472 model->setCompletion(completion);
Chris@0 473
Chris@0 474 } else if (binCount == 1) {
Chris@0 475
Chris@0 476 SparseTimeValueModel *model = getOutput<SparseTimeValueModel>();
Chris@0 477 if (!model) return;
Chris@0 478 model->setCompletion(completion);
Chris@0 479
Chris@0 480 } else if (m_descriptor->sampleType ==
Chris@0 481 Vamp::Plugin::OutputDescriptor::VariableSampleRate) {
Chris@0 482
Chris@0 483 NoteModel *model = getOutput<NoteModel>();
Chris@0 484 if (!model) return;
Chris@0 485 model->setCompletion(completion);
Chris@0 486
Chris@0 487 } else {
Chris@0 488
Chris@0 489 DenseThreeDimensionalModel *model = getOutput<DenseThreeDimensionalModel>();
Chris@0 490 if (!model) return;
Chris@0 491 model->setCompletion(completion);
Chris@0 492 }
Chris@0 493 }
Chris@0 494