annotate runner/FeatureExtractionManager.cpp @ 45:69c438d4b9d3

* Pick up default sample rate and channel count from first audio file (formerly they were hardcoded to 44100 and 1...)
author Chris Cannam
date Mon, 18 Oct 2010 14:17:48 +0100
parents 76ea8ffc4dfa
children 4d07f61dba3f
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 Annotator
Chris@0 5 A utility for batch feature extraction from audio files.
Chris@0 6 Mark Levy, Chris Sutton and Chris Cannam, Queen Mary, University of London.
Chris@0 7 Copyright 2007-2008 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@0 16 #include "FeatureExtractionManager.h"
Chris@0 17
Chris@0 18 #include <vamp-hostsdk/PluginChannelAdapter.h>
Chris@0 19 #include <vamp-hostsdk/PluginBufferingAdapter.h>
Chris@0 20 #include <vamp-hostsdk/PluginInputDomainAdapter.h>
Chris@0 21 #include <vamp-hostsdk/PluginSummarisingAdapter.h>
Chris@8 22 #include <vamp-hostsdk/PluginWrapper.h>
Chris@0 23 #include <vamp-hostsdk/PluginLoader.h>
Chris@0 24
Chris@21 25 #include "base/Exceptions.h"
Chris@21 26
Chris@0 27 #include <iostream>
Chris@0 28
Chris@0 29 using namespace std;
Chris@0 30
Chris@0 31 using Vamp::Plugin;
Chris@0 32 using Vamp::PluginBase;
Chris@0 33 using Vamp::HostExt::PluginLoader;
Chris@0 34 using Vamp::HostExt::PluginChannelAdapter;
Chris@0 35 using Vamp::HostExt::PluginBufferingAdapter;
Chris@0 36 using Vamp::HostExt::PluginInputDomainAdapter;
Chris@0 37 using Vamp::HostExt::PluginSummarisingAdapter;
Chris@8 38 using Vamp::HostExt::PluginWrapper;
Chris@0 39
Chris@0 40 #include "data/fileio/FileSource.h"
Chris@0 41 #include "data/fileio/AudioFileReader.h"
Chris@0 42 #include "data/fileio/AudioFileReaderFactory.h"
Chris@0 43 #include "data/fileio/PlaylistFileReader.h"
Chris@0 44 #include "base/TempDirectory.h"
Chris@0 45 #include "base/ProgressPrinter.h"
Chris@0 46 #include "transform/TransformFactory.h"
Chris@0 47 #include "rdf/RDFTransformFactory.h"
Chris@0 48 #include "transform/FeatureWriter.h"
Chris@0 49
Chris@0 50 #include <QTextStream>
Chris@0 51 #include <QFile>
Chris@0 52 #include <QFileInfo>
Chris@0 53
Chris@0 54 FeatureExtractionManager::FeatureExtractionManager() :
Chris@0 55 m_summariesOnly(false),
Chris@0 56 // We can read using an arbitrary fixed block size --
Chris@0 57 // PluginBufferingAdapter handles this for us. It's likely to be
Chris@0 58 // quicker to use larger sizes than smallish ones like 1024
Chris@0 59 m_blockSize(16384),
Chris@0 60 m_defaultSampleRate(0),
Chris@0 61 m_sampleRate(0),
Chris@45 62 m_channels(0)
Chris@0 63 {
Chris@0 64 }
Chris@0 65
Chris@0 66 FeatureExtractionManager::~FeatureExtractionManager()
Chris@0 67 {
Chris@0 68 for (PluginMap::iterator pi = m_plugins.begin();
Chris@0 69 pi != m_plugins.end(); ++pi) {
Chris@0 70 delete pi->first;
Chris@0 71 }
Chris@45 72 foreach (AudioFileReader *r, m_readyReaders) {
Chris@45 73 delete r;
Chris@45 74 }
Chris@0 75 }
Chris@0 76
Chris@0 77 void FeatureExtractionManager::setChannels(int channels)
Chris@0 78 {
Chris@0 79 m_channels = channels;
Chris@0 80 }
Chris@0 81
Chris@0 82 void FeatureExtractionManager::setDefaultSampleRate(int sampleRate)
Chris@0 83 {
Chris@0 84 m_defaultSampleRate = sampleRate;
Chris@0 85 }
Chris@0 86
Chris@0 87 static PluginSummarisingAdapter::SummaryType
Chris@0 88 getSummaryType(string name)
Chris@0 89 {
Chris@0 90 if (name == "min") return PluginSummarisingAdapter::Minimum;
Chris@0 91 if (name == "max") return PluginSummarisingAdapter::Maximum;
Chris@0 92 if (name == "mean") return PluginSummarisingAdapter::Mean;
Chris@0 93 if (name == "median") return PluginSummarisingAdapter::Median;
Chris@0 94 if (name == "mode") return PluginSummarisingAdapter::Mode;
Chris@0 95 if (name == "sum") return PluginSummarisingAdapter::Sum;
Chris@0 96 if (name == "variance") return PluginSummarisingAdapter::Variance;
Chris@0 97 if (name == "sd") return PluginSummarisingAdapter::StandardDeviation;
Chris@0 98 if (name == "count") return PluginSummarisingAdapter::Count;
Chris@0 99 return PluginSummarisingAdapter::UnknownSummaryType;
Chris@0 100 }
Chris@0 101
Chris@0 102 bool FeatureExtractionManager::setSummaryTypes(const set<string> &names,
Chris@0 103 bool summariesOnly,
Chris@0 104 const PluginSummarisingAdapter::SegmentBoundaries &boundaries)
Chris@0 105 {
Chris@0 106 for (SummaryNameSet::const_iterator i = names.begin();
Chris@0 107 i != names.end(); ++i) {
Chris@0 108 if (getSummaryType(*i) == PluginSummarisingAdapter::UnknownSummaryType) {
Chris@0 109 cerr << "ERROR: Unknown summary type \"" << *i << "\"" << endl;
Chris@0 110 return false;
Chris@0 111 }
Chris@0 112 }
Chris@0 113 m_summaries = names;
Chris@0 114 m_summariesOnly = summariesOnly;
Chris@0 115 m_boundaries = boundaries;
Chris@0 116 return true;
Chris@0 117 }
Chris@0 118
Chris@0 119 bool FeatureExtractionManager::addFeatureExtractor
Chris@0 120 (Transform transform, const vector<FeatureWriter*> &writers)
Chris@0 121 {
Chris@0 122 //!!! exceptions rather than return values?
Chris@0 123
Chris@0 124 if (transform.getSampleRate() == 0) {
Chris@0 125 if (m_sampleRate == 0) {
Chris@0 126 cerr << "NOTE: Transform does not specify a sample rate, using default rate of " << m_defaultSampleRate << endl;
Chris@0 127 transform.setSampleRate(m_defaultSampleRate);
Chris@0 128 m_sampleRate = m_defaultSampleRate;
Chris@0 129 } else {
Chris@0 130 cerr << "NOTE: Transform does not specify a sample rate, using previous transform's rate of " << m_sampleRate << endl;
Chris@0 131 transform.setSampleRate(m_sampleRate);
Chris@0 132 }
Chris@0 133 }
Chris@0 134
Chris@0 135 if (m_sampleRate == 0) {
Chris@0 136 m_sampleRate = transform.getSampleRate();
Chris@0 137 }
Chris@0 138
Chris@0 139 if (transform.getSampleRate() != m_sampleRate) {
Chris@0 140 cerr << "WARNING: Transform sample rate " << transform.getSampleRate() << " does not match previously specified transform rate of " << m_sampleRate << " -- only a single rate is supported for each run" << endl;
Chris@0 141 cerr << "WARNING: Using previous rate of " << m_sampleRate << " for this transform as well" << endl;
Chris@0 142 transform.setSampleRate(m_sampleRate);
Chris@0 143 }
Chris@0 144
Chris@0 145 Plugin *plugin = 0;
Chris@0 146
Chris@0 147 // Remember what the original transform looked like, and index
Chris@0 148 // based on this -- because we may be about to fill in the zeros
Chris@0 149 // for step and block size, but we want any further copies with
Chris@0 150 // the same zeros to match this one
Chris@0 151 Transform originalTransform = transform;
Chris@0 152
Chris@0 153 if (m_transformPluginMap.find(transform) == m_transformPluginMap.end()) {
Chris@0 154
Chris@0 155 // Test whether we already have a transform that is identical
Chris@0 156 // to this, except for the output requested and/or the summary
Chris@0 157 // type -- if so, they should share plugin instances (a vital
Chris@0 158 // optimisation)
Chris@0 159
Chris@0 160 for (TransformPluginMap::iterator i = m_transformPluginMap.begin();
Chris@0 161 i != m_transformPluginMap.end(); ++i) {
Chris@0 162 Transform test = i->first;
Chris@0 163 test.setOutput(transform.getOutput());
Chris@0 164 test.setSummaryType(transform.getSummaryType());
Chris@0 165 if (transform == test) {
Chris@0 166 cerr << "NOTE: Already have transform identical to this one (for \""
Chris@0 167 << transform.getIdentifier().toStdString()
Chris@0 168 << "\") in every detail except output identifier and/or "
Chris@0 169 << "summary type; sharing its plugin instance" << endl;
Chris@0 170 plugin = i->second;
Chris@0 171 if (transform.getSummaryType() != Transform::NoSummary &&
Chris@0 172 !dynamic_cast<PluginSummarisingAdapter *>(plugin)) {
Chris@0 173 plugin = new PluginSummarisingAdapter(plugin);
Chris@0 174 i->second = plugin;
Chris@0 175 }
Chris@0 176 break;
Chris@0 177 }
Chris@0 178 }
Chris@0 179
Chris@0 180 if (!plugin) {
Chris@0 181
Chris@0 182 TransformFactory *tf = TransformFactory::getInstance();
Chris@0 183
Chris@0 184 PluginBase *pb = tf->instantiatePluginFor(transform);
Chris@0 185 plugin = tf->downcastVampPlugin(pb);
Chris@0 186 if (!plugin) {
Chris@0 187 //!!! todo: handle non-Vamp plugins too, or make the main --list
Chris@0 188 // option print out only Vamp transforms
Chris@0 189 cerr << "ERROR: Failed to load plugin for transform \""
Chris@0 190 << transform.getIdentifier().toStdString() << "\"" << endl;
Chris@0 191 delete pb;
Chris@0 192 return false;
Chris@0 193 }
Chris@0 194
Chris@0 195 // We will provide the plugin with arbitrary step and
Chris@0 196 // block sizes (so that we can use the same read/write
Chris@0 197 // block size for all transforms), and to that end we use
Chris@0 198 // a PluginBufferingAdapter. However, we need to know the
Chris@0 199 // underlying step size so that we can provide the right
Chris@0 200 // context for dense outputs. (Although, don't forget
Chris@0 201 // that the PluginBufferingAdapter rewrites
Chris@0 202 // OneSamplePerStep outputs so as to use FixedSampleRate
Chris@0 203 // -- so it supplies the sample rate in the output
Chris@0 204 // feature. I'm not sure whether we can easily use that.)
Chris@0 205
Chris@0 206 size_t pluginStepSize = plugin->getPreferredStepSize();
Chris@0 207 size_t pluginBlockSize = plugin->getPreferredBlockSize();
Chris@0 208
Chris@25 209 PluginInputDomainAdapter *pida = 0;
Chris@25 210
Chris@0 211 // adapt the plugin for buffering, channels, etc.
Chris@0 212 if (plugin->getInputDomain() == Plugin::FrequencyDomain) {
Chris@25 213 pida = new PluginInputDomainAdapter(plugin);
Chris@26 214 pida->setProcessTimestampMethod(PluginInputDomainAdapter::ShiftData);
Chris@25 215 plugin = pida;
Chris@0 216 }
Chris@0 217
Chris@0 218 PluginBufferingAdapter *pba = new PluginBufferingAdapter(plugin);
Chris@0 219 plugin = pba;
Chris@0 220
Chris@0 221 if (transform.getStepSize() != 0) {
Chris@0 222 pba->setPluginStepSize(transform.getStepSize());
Chris@0 223 } else {
Chris@0 224 transform.setStepSize(pluginStepSize);
Chris@0 225 }
Chris@0 226
Chris@0 227 if (transform.getBlockSize() != 0) {
Chris@0 228 pba->setPluginBlockSize(transform.getBlockSize());
Chris@0 229 } else {
Chris@0 230 transform.setBlockSize(pluginBlockSize);
Chris@0 231 }
Chris@0 232
Chris@0 233 plugin = new PluginChannelAdapter(plugin);
Chris@0 234
Chris@0 235 if (!m_summaries.empty() ||
Chris@0 236 transform.getSummaryType() != Transform::NoSummary) {
Chris@0 237 PluginSummarisingAdapter *adapter =
Chris@0 238 new PluginSummarisingAdapter(plugin);
Chris@0 239 adapter->setSummarySegmentBoundaries(m_boundaries);
Chris@0 240 plugin = adapter;
Chris@0 241 }
Chris@0 242
Chris@0 243 if (!plugin->initialise(m_channels, m_blockSize, m_blockSize)) {
Chris@0 244 cerr << "ERROR: Plugin initialise (channels = " << m_channels << ", stepSize = " << m_blockSize << ", blockSize = " << m_blockSize << ") failed." << endl;
Chris@0 245 delete plugin;
Chris@0 246 return false;
Chris@0 247 }
Chris@0 248
Chris@0 249 // cerr << "Initialised plugin" << endl;
Chris@0 250
Chris@0 251 size_t actualStepSize = 0;
Chris@0 252 size_t actualBlockSize = 0;
Chris@0 253 pba->getActualStepAndBlockSizes(actualStepSize, actualBlockSize);
Chris@0 254 transform.setStepSize(actualStepSize);
Chris@0 255 transform.setBlockSize(actualBlockSize);
Chris@0 256
Chris@0 257 Plugin::OutputList outputs = plugin->getOutputDescriptors();
Chris@0 258 for (int i = 0; i < (int)outputs.size(); ++i) {
Chris@0 259
Chris@0 260 // cerr << "Newly initialised plugin output " << i << " has bin count " << outputs[i].binCount << endl;
Chris@0 261
Chris@0 262 m_pluginOutputs[plugin][outputs[i].identifier] = outputs[i];
Chris@0 263 m_pluginOutputIndices[outputs[i].identifier] = i;
Chris@0 264 }
Chris@0 265
Chris@10 266 cerr << "NOTE: Loaded and initialised plugin for transform \""
Chris@25 267 << transform.getIdentifier().toStdString()
Chris@25 268 << "\" with plugin step size " << actualStepSize
Chris@25 269 << " and block size " << actualBlockSize
Chris@25 270 << " (adapter step and block size " << m_blockSize << ")"
Chris@25 271 << endl;
Chris@25 272
Chris@25 273 if (pida) {
Chris@25 274 cerr << "NOTE: PluginInputDomainAdapter timestamp adjustment is "
Chris@25 275
Chris@25 276 << pida->getTimestampAdjustment() << endl;
Chris@25 277 }
Chris@8 278
Chris@8 279 } else {
Chris@8 280
Chris@8 281 if (transform.getStepSize() == 0 || transform.getBlockSize() == 0) {
Chris@8 282
Chris@8 283 PluginWrapper *pw = dynamic_cast<PluginWrapper *>(plugin);
Chris@8 284 if (pw) {
Chris@8 285 PluginBufferingAdapter *pba =
Chris@8 286 pw->getWrapper<PluginBufferingAdapter>();
Chris@8 287 if (pba) {
Chris@8 288 size_t actualStepSize = 0;
Chris@8 289 size_t actualBlockSize = 0;
Chris@8 290 pba->getActualStepAndBlockSizes(actualStepSize,
Chris@8 291 actualBlockSize);
Chris@8 292 if (transform.getStepSize() == 0) {
Chris@8 293 transform.setStepSize(actualStepSize);
Chris@8 294 }
Chris@8 295 if (transform.getBlockSize() == 0) {
Chris@8 296 transform.setBlockSize(actualBlockSize);
Chris@8 297 }
Chris@8 298 }
Chris@8 299 }
Chris@8 300 }
Chris@0 301 }
Chris@0 302
Chris@0 303 if (transform.getOutput() == "") {
Chris@0 304 transform.setOutput
Chris@0 305 (plugin->getOutputDescriptors()[0].identifier.c_str());
Chris@0 306 }
Chris@0 307
Chris@0 308 m_transformPluginMap[transform] = plugin;
Chris@0 309
Chris@0 310 if (!(originalTransform == transform)) {
Chris@0 311 m_transformPluginMap[originalTransform] = plugin;
Chris@0 312 }
Chris@0 313
Chris@0 314 } else {
Chris@0 315
Chris@0 316 plugin = m_transformPluginMap[transform];
Chris@0 317 }
Chris@0 318
Chris@0 319 m_plugins[plugin][transform] = writers;
Chris@0 320
Chris@0 321 return true;
Chris@0 322 }
Chris@0 323
Chris@0 324 bool FeatureExtractionManager::addDefaultFeatureExtractor
Chris@0 325 (TransformId transformId, const vector<FeatureWriter*> &writers)
Chris@0 326 {
Chris@0 327 TransformFactory *tf = TransformFactory::getInstance();
Chris@0 328
Chris@0 329 if (m_sampleRate == 0) {
Chris@0 330 if (m_defaultSampleRate == 0) {
Chris@0 331 cerr << "ERROR: Default transform requested, but no default sample rate available" << endl;
Chris@0 332 return false;
Chris@0 333 } else {
Chris@0 334 cerr << "NOTE: Using default sample rate of " << m_defaultSampleRate << " for default transform" << endl;
Chris@0 335 m_sampleRate = m_defaultSampleRate;
Chris@0 336 }
Chris@0 337 }
Chris@0 338
Chris@0 339 Transform transform = tf->getDefaultTransformFor(transformId, m_sampleRate);
Chris@0 340
Chris@0 341 return addFeatureExtractor(transform, writers);
Chris@0 342 }
Chris@0 343
Chris@0 344 bool FeatureExtractionManager::addFeatureExtractorFromFile
Chris@0 345 (QString transformXmlFile, const vector<FeatureWriter*> &writers)
Chris@0 346 {
Chris@0 347 RDFTransformFactory factory
Chris@0 348 (QUrl::fromLocalFile(QFileInfo(transformXmlFile).absoluteFilePath())
Chris@0 349 .toString());
Chris@0 350 ProgressPrinter printer("Parsing transforms RDF file");
Chris@0 351 std::vector<Transform> transforms = factory.getTransforms(&printer);
Chris@0 352 if (!factory.isOK()) {
Chris@0 353 cerr << "WARNING: FeatureExtractionManager::addFeatureExtractorFromFile: Failed to parse transforms file: " << factory.getErrorString().toStdString() << endl;
Chris@0 354 if (factory.isRDF()) {
Chris@0 355 return false; // no point trying it as XML
Chris@0 356 }
Chris@0 357 }
Chris@0 358 if (!transforms.empty()) {
Chris@0 359 bool success = true;
Chris@0 360 for (int i = 0; i < (int)transforms.size(); ++i) {
Chris@0 361 if (!addFeatureExtractor(transforms[i], writers)) {
Chris@0 362 success = false;
Chris@0 363 }
Chris@0 364 }
Chris@0 365 return success;
Chris@0 366 }
Chris@0 367
Chris@0 368 QFile file(transformXmlFile);
Chris@0 369 if (!file.open(QIODevice::ReadOnly | QIODevice::Text)) {
Chris@0 370 cerr << "ERROR: Failed to open transform XML file \""
Chris@0 371 << transformXmlFile.toStdString() << "\" for reading" << endl;
Chris@0 372 return false;
Chris@0 373 }
Chris@0 374
Chris@0 375 QTextStream *qts = new QTextStream(&file);
Chris@0 376 QString qs = qts->readAll();
Chris@0 377 delete qts;
Chris@0 378 file.close();
Chris@0 379
Chris@0 380 Transform transform(qs);
Chris@0 381
Chris@0 382 return addFeatureExtractor(transform, writers);
Chris@0 383 }
Chris@0 384
Chris@45 385 void FeatureExtractionManager::addSource(QString audioSource)
Chris@0 386 {
Chris@0 387 if (QFileInfo(audioSource).suffix().toLower() == "m3u") {
Chris@45 388 ProgressPrinter retrievalProgress("Opening playlist file...");
Chris@45 389 FileSource source(audioSource, &retrievalProgress);
Chris@45 390 if (!source.isAvailable()) {
Chris@45 391 cerr << "ERROR: File or URL \"" << audioSource.toStdString()
Chris@45 392 << "\" could not be located" << endl;
Chris@45 393 throw FileNotFound(audioSource);
Chris@45 394 }
Chris@45 395 source.waitForData();
Chris@0 396 PlaylistFileReader reader(source);
Chris@0 397 if (reader.isOK()) {
Chris@0 398 vector<QString> files = reader.load();
Chris@0 399 for (int i = 0; i < (int)files.size(); ++i) {
Chris@45 400 addSource(files[i]);
Chris@0 401 }
Chris@0 402 return;
Chris@0 403 } else {
Chris@0 404 cerr << "ERROR: Playlist \"" << audioSource.toStdString()
Chris@0 405 << "\" could not be opened" << endl;
Chris@21 406 throw FileNotFound(audioSource);
Chris@0 407 }
Chris@0 408 }
Chris@0 409
Chris@45 410 std::cerr << "Have audio source: \"" << audioSource.toStdString() << "\"" << std::endl;
Chris@45 411
Chris@45 412 // We don't actually do anything with it here, unless it's the
Chris@45 413 // first audio source and we need it to establish default channel
Chris@45 414 // count and sample rate
Chris@45 415
Chris@45 416 if (m_channels == 0 || m_defaultSampleRate == 0) {
Chris@45 417
Chris@45 418 ProgressPrinter retrievalProgress("Determining default rate and channel count from first input file...");
Chris@45 419
Chris@45 420 FileSource source(audioSource, &retrievalProgress);
Chris@45 421 if (!source.isAvailable()) {
Chris@45 422 cerr << "ERROR: File or URL \"" << audioSource.toStdString()
Chris@45 423 << "\" could not be located" << endl;
Chris@45 424 throw FileNotFound(audioSource);
Chris@45 425 }
Chris@45 426
Chris@45 427 source.waitForData();
Chris@45 428
Chris@45 429 // Open to determine validity, channel count, sample rate only
Chris@45 430 // (then close, and open again later with actual desired rate &c)
Chris@45 431
Chris@45 432 AudioFileReader *reader =
Chris@45 433 AudioFileReaderFactory::createReader(source, 0, &retrievalProgress);
Chris@45 434
Chris@45 435 if (!reader) {
Chris@45 436 throw FailedToOpenFile(audioSource);
Chris@45 437 }
Chris@45 438
Chris@45 439 retrievalProgress.done();
Chris@45 440
Chris@45 441 cerr << "File or URL \"" << audioSource.toStdString() << "\" opened successfully" << endl;
Chris@45 442
Chris@45 443 if (m_channels == 0) {
Chris@45 444 m_channels = reader->getChannelCount();
Chris@45 445 cerr << "Taking default channel count of "
Chris@45 446 << reader->getChannelCount() << " from file" << endl;
Chris@45 447 }
Chris@45 448
Chris@45 449 if (m_defaultSampleRate == 0) {
Chris@45 450 m_defaultSampleRate = reader->getNativeRate();
Chris@45 451 cerr << "Taking default sample rate of "
Chris@45 452 << reader->getNativeRate() << "Hz from file" << endl;
Chris@45 453 cerr << "(Note: Default may be overridden by transforms)" << endl;
Chris@45 454 }
Chris@45 455
Chris@45 456 m_readyReaders[audioSource] = reader;
Chris@45 457 }
Chris@45 458 }
Chris@45 459
Chris@45 460 void FeatureExtractionManager::extractFeatures(QString audioSource)
Chris@45 461 {
Chris@45 462 if (m_plugins.empty()) return;
Chris@45 463
Chris@45 464 testOutputFiles(audioSource);
Chris@45 465
Chris@0 466 if (m_sampleRate == 0) {
Chris@45 467 throw FileOperationFailed
Chris@45 468 (audioSource, "internal error: have sources and plugins, but no sample rate");
Chris@45 469 }
Chris@45 470 if (m_channels == 0) {
Chris@45 471 throw FileOperationFailed
Chris@45 472 (audioSource, "internal error: have sources and plugins, but no channel count");
Chris@0 473 }
Chris@0 474
Chris@45 475 AudioFileReader *reader = 0;
Chris@45 476
Chris@45 477 if (m_readyReaders.contains(audioSource)) {
Chris@45 478 reader = m_readyReaders[audioSource];
Chris@45 479 m_readyReaders.remove(audioSource);
Chris@45 480 if (reader->getChannelCount() != m_channels ||
Chris@45 481 reader->getSampleRate() != m_sampleRate) {
Chris@45 482 // can't use this; open it again
Chris@45 483 delete reader;
Chris@45 484 reader = 0;
Chris@45 485 }
Chris@45 486 }
Chris@45 487 if (!reader) {
Chris@45 488 ProgressPrinter retrievalProgress("Retrieving audio data...");
Chris@45 489 FileSource source(audioSource, &retrievalProgress);
Chris@45 490 source.waitForData();
Chris@45 491 reader = AudioFileReaderFactory::createReader
Chris@45 492 (source, m_sampleRate, &retrievalProgress);
Chris@45 493 retrievalProgress.done();
Chris@45 494 }
Chris@45 495
Chris@0 496 if (!reader) {
Chris@21 497 throw FailedToOpenFile(audioSource);
Chris@0 498 }
Chris@0 499
Chris@45 500 cerr << "Audio file \"" << audioSource.toStdString() << "\": "
Chris@45 501 << reader->getChannelCount() << "ch at "
Chris@45 502 << reader->getNativeRate() << "Hz" << endl;
Chris@45 503 if (reader->getChannelCount() != m_channels ||
Chris@45 504 reader->getNativeRate() != m_sampleRate) {
Chris@45 505 cerr << "NOTE: File will be mixed or resampled for processing: "
Chris@45 506 << m_channels << "ch at "
Chris@45 507 << m_sampleRate << "Hz" << endl;
Chris@45 508 }
Chris@11 509
Chris@0 510 // allocate audio buffers
Chris@0 511 float **data = new float *[m_channels];
Chris@0 512 for (int c = 0; c < m_channels; ++c) {
Chris@0 513 data[c] = new float[m_blockSize];
Chris@0 514 }
Chris@31 515
Chris@31 516 struct LifespanMgr { // unintrusive hack introduced to ensure
Chris@31 517 // destruction on exceptions
Chris@31 518 AudioFileReader *m_r;
Chris@31 519 int m_c;
Chris@31 520 float **m_d;
Chris@31 521 LifespanMgr(AudioFileReader *r, int c, float **d) :
Chris@31 522 m_r(r), m_c(c), m_d(d) { }
Chris@31 523 ~LifespanMgr() { destroy(); }
Chris@31 524 void destroy() {
Chris@31 525 if (!m_r) return;
Chris@31 526 delete m_r;
Chris@31 527 for (int i = 0; i < m_c; ++i) delete[] m_d[i];
Chris@31 528 delete[] m_d;
Chris@31 529 m_r = 0;
Chris@31 530 }
Chris@31 531 };
Chris@31 532 LifespanMgr lifemgr(reader, m_channels, data);
Chris@0 533
Chris@0 534 size_t frameCount = reader->getFrameCount();
Chris@0 535
Chris@0 536 // cerr << "file has " << frameCount << " frames" << endl;
Chris@0 537
Chris@0 538 for (PluginMap::iterator pi = m_plugins.begin();
Chris@0 539 pi != m_plugins.end(); ++pi) {
Chris@0 540
Chris@0 541 Plugin *plugin = pi->first;
Chris@0 542
Chris@0 543 // std::cerr << "Calling reset on " << plugin << std::endl;
Chris@0 544 plugin->reset();
Chris@0 545
Chris@0 546 for (TransformWriterMap::iterator ti = pi->second.begin();
Chris@0 547 ti != pi->second.end(); ++ti) {
Chris@0 548
Chris@0 549 const Transform &transform = ti->first;
Chris@0 550
Chris@0 551 //!!! we may want to set the start and duration times for extraction
Chris@0 552 // in the transform record (defaults of zero indicate extraction
Chris@0 553 // from the whole file)
Chris@0 554 // transform.setStartTime(RealTime::zeroTime);
Chris@0 555 // transform.setDuration
Chris@0 556 // (RealTime::frame2RealTime(reader->getFrameCount(), m_sampleRate));
Chris@0 557
Chris@0 558 string outputId = transform.getOutput().toStdString();
Chris@0 559 if (m_pluginOutputs[plugin].find(outputId) ==
Chris@0 560 m_pluginOutputs[plugin].end()) {
Chris@0 561 //!!! throw?
Chris@0 562 cerr << "WARNING: Nonexistent plugin output \"" << outputId << "\" requested for transform \""
Chris@0 563 << transform.getIdentifier().toStdString() << "\", ignoring this transform"
Chris@0 564 << endl;
Chris@0 565 /*
Chris@0 566 cerr << "Known outputs for all plugins are as follows:" << endl;
Chris@0 567 for (PluginOutputMap::const_iterator k = m_pluginOutputs.begin();
Chris@0 568 k != m_pluginOutputs.end(); ++k) {
Chris@0 569 cerr << "Plugin " << k->first << ": ";
Chris@0 570 if (k->second.empty()) {
Chris@0 571 cerr << "(none)";
Chris@0 572 }
Chris@0 573 for (OutputMap::const_iterator i = k->second.begin();
Chris@0 574 i != k->second.end(); ++i) {
Chris@0 575 cerr << "\"" << i->first << "\" ";
Chris@0 576 }
Chris@0 577 cerr << endl;
Chris@0 578 }
Chris@0 579 */
Chris@0 580 }
Chris@0 581 }
Chris@0 582 }
Chris@0 583
Chris@0 584 long startFrame = 0;
Chris@0 585 long endFrame = frameCount;
Chris@0 586
Chris@0 587 /*!!! No -- there is no single transform to pull this stuff from --
Chris@0 588 * the transforms may have various start and end times, need to be far
Chris@0 589 * cleverer about this if we're going to support them
Chris@0 590
Chris@0 591 RealTime trStartRT = transform.getStartTime();
Chris@0 592 RealTime trDurationRT = transform.getDuration();
Chris@0 593
Chris@0 594 long trStart = RealTime::realTime2Frame(trStartRT, m_sampleRate);
Chris@0 595 long trDuration = RealTime::realTime2Frame(trDurationRT, m_sampleRate);
Chris@0 596
Chris@0 597 if (trStart == 0 || trStart < startFrame) {
Chris@0 598 trStart = startFrame;
Chris@0 599 }
Chris@0 600
Chris@0 601 if (trDuration == 0) {
Chris@0 602 trDuration = endFrame - trStart;
Chris@0 603 }
Chris@0 604 if (trStart + trDuration > endFrame) {
Chris@0 605 trDuration = endFrame - trStart;
Chris@0 606 }
Chris@0 607
Chris@0 608 startFrame = trStart;
Chris@0 609 endFrame = trStart + trDuration;
Chris@0 610 */
Chris@0 611
Chris@0 612 for (PluginMap::iterator pi = m_plugins.begin();
Chris@0 613 pi != m_plugins.end(); ++pi) {
Chris@0 614
Chris@0 615 for (TransformWriterMap::const_iterator ti = pi->second.begin();
Chris@0 616 ti != pi->second.end(); ++ti) {
Chris@0 617
Chris@0 618 const vector<FeatureWriter *> &writers = ti->second;
Chris@0 619
Chris@0 620 for (int j = 0; j < (int)writers.size(); ++j) {
Chris@0 621 FeatureWriter::TrackMetadata m;
Chris@0 622 m.title = reader->getTitle();
Chris@0 623 m.maker = reader->getMaker();
Chris@19 624 if (m.title != "" && m.maker != "") {
Chris@19 625 writers[j]->setTrackMetadata(audioSource, m);
Chris@19 626 }
Chris@0 627 }
Chris@0 628 }
Chris@0 629 }
Chris@0 630
Chris@0 631 ProgressPrinter extractionProgress("Extracting and writing features...");
Chris@0 632 int progress = 0;
Chris@0 633
Chris@0 634 for (long i = startFrame; i < endFrame; i += m_blockSize) {
Chris@0 635
Chris@0 636 //!!! inefficient, although much of the inefficiency may be
Chris@0 637 // susceptible to optimisation
Chris@0 638
Chris@0 639 SampleBlock frames;
Chris@0 640 reader->getInterleavedFrames(i, m_blockSize, frames);
Chris@0 641
Chris@0 642 // We have to do our own channel handling here; we can't just
Chris@0 643 // leave it to the plugin adapter because the same plugin
Chris@0 644 // adapter may have to serve for input files with various
Chris@0 645 // numbers of channels (so the adapter is simply configured
Chris@34 646 // with a fixed channel count).
Chris@0 647
Chris@0 648 int rc = reader->getChannelCount();
Chris@0 649
Chris@34 650 // m_channels is the number of channels we need for the plugin
Chris@34 651
Chris@34 652 int index;
Chris@34 653 int fc = (int)frames.size();
Chris@34 654 if (m_channels == 1) { // only case in which we can sensibly mix down
Chris@34 655 for (int j = 0; j < m_blockSize; ++j) {
Chris@34 656 data[0][j] = 0.f;
Chris@34 657 }
Chris@34 658 for (int c = 0; c < rc; ++c) {
Chris@34 659 for (int j = 0; j < m_blockSize; ++j) {
Chris@0 660 index = j * rc + c;
Chris@34 661 if (index < fc) data[0][j] += frames[index];
Chris@0 662 }
Chris@0 663 }
Chris@34 664 for (int j = 0; j < m_blockSize; ++j) {
Chris@34 665 data[0][j] /= rc;
Chris@34 666 }
Chris@34 667 } else {
Chris@34 668 for (int c = 0; c < m_channels; ++c) {
Chris@34 669 for (int j = 0; j < m_blockSize; ++j) {
Chris@34 670 data[c][j] = 0.f;
Chris@34 671 }
Chris@34 672 if (c < rc) {
Chris@34 673 for (int j = 0; j < m_blockSize; ++j) {
Chris@34 674 index = j * rc + c;
Chris@34 675 if (index < fc) data[c][j] += frames[index];
Chris@34 676 }
Chris@34 677 }
Chris@34 678 }
Chris@34 679 }
Chris@0 680
Chris@0 681 Vamp::RealTime timestamp = Vamp::RealTime::frame2RealTime
Chris@0 682 (i, m_sampleRate);
Chris@0 683
Chris@0 684 for (PluginMap::iterator pi = m_plugins.begin();
Chris@0 685 pi != m_plugins.end(); ++pi) {
Chris@0 686
Chris@0 687 Plugin *plugin = pi->first;
Chris@0 688 Plugin::FeatureSet featureSet = plugin->process(data, timestamp);
Chris@0 689
Chris@0 690 if (!m_summariesOnly) {
Chris@0 691 writeFeatures(audioSource, plugin, featureSet);
Chris@0 692 }
Chris@0 693 }
Chris@0 694
Chris@0 695 int pp = progress;
Chris@6 696 progress = int(((i - startFrame) * 100.0) / (endFrame - startFrame) + 0.1);
Chris@0 697 if (progress > pp) extractionProgress.setProgress(progress);
Chris@0 698 }
Chris@10 699
Chris@22 700 // std::cerr << "FeatureExtractionManager: deleting audio file reader" << std::endl;
Chris@12 701
Chris@31 702 lifemgr.destroy(); // deletes reader, data
Chris@0 703
Chris@0 704 for (PluginMap::iterator pi = m_plugins.begin();
Chris@0 705 pi != m_plugins.end(); ++pi) {
Chris@0 706
Chris@0 707 Plugin *plugin = pi->first;
Chris@0 708 Plugin::FeatureSet featureSet = plugin->getRemainingFeatures();
Chris@0 709
Chris@0 710 if (!m_summariesOnly) {
Chris@0 711 writeFeatures(audioSource, plugin, featureSet);
Chris@0 712 }
Chris@0 713
Chris@0 714 if (!m_summaries.empty()) {
Chris@0 715 PluginSummarisingAdapter *adapter =
Chris@0 716 dynamic_cast<PluginSummarisingAdapter *>(plugin);
Chris@0 717 if (!adapter) {
Chris@0 718 cerr << "WARNING: Summaries requested, but plugin is not a summarising adapter" << endl;
Chris@0 719 } else {
Chris@0 720 for (SummaryNameSet::const_iterator sni = m_summaries.begin();
Chris@0 721 sni != m_summaries.end(); ++sni) {
Chris@0 722 featureSet.clear();
Chris@0 723 //!!! problem here -- we are requesting summaries
Chris@0 724 //!!! for all outputs, but they in principle have
Chris@0 725 //!!! different averaging requirements depending
Chris@0 726 //!!! on whether their features have duration or
Chris@0 727 //!!! not
Chris@0 728 featureSet = adapter->getSummaryForAllOutputs
Chris@0 729 (getSummaryType(*sni),
Chris@0 730 PluginSummarisingAdapter::ContinuousTimeAverage);
Chris@0 731 writeFeatures(audioSource, plugin, featureSet,//!!! *sni);
Chris@0 732 Transform::stringToSummaryType(sni->c_str()));
Chris@0 733 }
Chris@0 734 }
Chris@0 735 }
Chris@0 736
Chris@0 737 writeSummaries(audioSource, plugin);
Chris@0 738 }
Chris@0 739
Chris@3 740 extractionProgress.done();
Chris@3 741
Chris@0 742 finish();
Chris@0 743
Chris@0 744 TempDirectory::getInstance()->cleanup();
Chris@0 745 }
Chris@0 746
Chris@0 747 void
Chris@0 748 FeatureExtractionManager::writeSummaries(QString audioSource, Plugin *plugin)
Chris@0 749 {
Chris@0 750 // caller should have ensured plugin is in m_plugins
Chris@0 751 PluginMap::iterator pi = m_plugins.find(plugin);
Chris@0 752
Chris@0 753 for (TransformWriterMap::const_iterator ti = pi->second.begin();
Chris@0 754 ti != pi->second.end(); ++ti) {
Chris@0 755
Chris@0 756 const Transform &transform = ti->first;
Chris@0 757 const vector<FeatureWriter *> &writers = ti->second;
Chris@0 758
Chris@0 759 Transform::SummaryType summaryType = transform.getSummaryType();
Chris@0 760 PluginSummarisingAdapter::SummaryType pType =
Chris@0 761 (PluginSummarisingAdapter::SummaryType)summaryType;
Chris@0 762
Chris@0 763 if (transform.getSummaryType() == Transform::NoSummary) {
Chris@0 764 continue;
Chris@0 765 }
Chris@0 766
Chris@0 767 PluginSummarisingAdapter *adapter =
Chris@0 768 dynamic_cast<PluginSummarisingAdapter *>(plugin);
Chris@0 769 if (!adapter) {
Chris@0 770 cerr << "FeatureExtractionManager::writeSummaries: INTERNAL ERROR: Summary requested for transform, but plugin is not a summarising adapter" << endl;
Chris@0 771 continue;
Chris@0 772 }
Chris@0 773
Chris@0 774 Plugin::FeatureSet featureSet = adapter->getSummaryForAllOutputs
Chris@0 775 (pType, PluginSummarisingAdapter::ContinuousTimeAverage);
Chris@0 776
Chris@0 777 // cout << "summary type " << int(pType) << " for transform:" << endl << transform.toXmlString().toStdString()<< endl << "... feature set with " << featureSet.size() << " elts" << endl;
Chris@0 778
Chris@0 779 writeFeatures(audioSource, plugin, featureSet, summaryType);
Chris@0 780 }
Chris@0 781 }
Chris@0 782
Chris@0 783 void FeatureExtractionManager::writeFeatures(QString audioSource,
Chris@0 784 Plugin *plugin,
Chris@0 785 const Plugin::FeatureSet &features,
Chris@0 786 Transform::SummaryType summaryType)
Chris@0 787 {
Chris@0 788 // caller should have ensured plugin is in m_plugins
Chris@0 789 PluginMap::iterator pi = m_plugins.find(plugin);
Chris@0 790
Chris@0 791 for (TransformWriterMap::const_iterator ti = pi->second.begin();
Chris@0 792 ti != pi->second.end(); ++ti) {
Chris@0 793
Chris@0 794 const Transform &transform = ti->first;
Chris@0 795 const vector<FeatureWriter *> &writers = ti->second;
Chris@0 796
Chris@0 797 if (transform.getSummaryType() != Transform::NoSummary &&
Chris@0 798 m_summaries.empty() &&
Chris@0 799 summaryType == Transform::NoSummary) {
Chris@0 800 continue;
Chris@0 801 }
Chris@0 802
Chris@0 803 if (transform.getSummaryType() != Transform::NoSummary &&
Chris@0 804 summaryType != Transform::NoSummary &&
Chris@0 805 transform.getSummaryType() != summaryType) {
Chris@0 806 continue;
Chris@0 807 }
Chris@0 808
Chris@0 809 string outputId = transform.getOutput().toStdString();
Chris@0 810
Chris@0 811 if (m_pluginOutputs[plugin].find(outputId) ==
Chris@0 812 m_pluginOutputs[plugin].end()) {
Chris@0 813 continue;
Chris@0 814 }
Chris@0 815
Chris@0 816 const Plugin::OutputDescriptor &desc =
Chris@0 817 m_pluginOutputs[plugin][outputId];
Chris@0 818
Chris@0 819 int outputIndex = m_pluginOutputIndices[outputId];
Chris@0 820 Plugin::FeatureSet::const_iterator fsi = features.find(outputIndex);
Chris@0 821 if (fsi == features.end()) continue;
Chris@0 822
Chris@0 823 for (int j = 0; j < (int)writers.size(); ++j) {
Chris@0 824 writers[j]->write
Chris@0 825 (audioSource, transform, desc, fsi->second,
Chris@0 826 Transform::summaryTypeToString(summaryType).toStdString());
Chris@0 827 }
Chris@0 828 }
Chris@0 829 }
Chris@0 830
Chris@31 831 void FeatureExtractionManager::testOutputFiles(QString audioSource)
Chris@31 832 {
Chris@31 833 for (PluginMap::iterator pi = m_plugins.begin();
Chris@31 834 pi != m_plugins.end(); ++pi) {
Chris@31 835
Chris@31 836 for (TransformWriterMap::iterator ti = pi->second.begin();
Chris@31 837 ti != pi->second.end(); ++ti) {
Chris@31 838
Chris@31 839 vector<FeatureWriter *> &writers = ti->second;
Chris@31 840
Chris@31 841 for (int i = 0; i < (int)writers.size(); ++i) {
Chris@31 842 writers[i]->testOutputFile(audioSource, ti->first.getIdentifier());
Chris@31 843 }
Chris@31 844 }
Chris@31 845 }
Chris@31 846 }
Chris@31 847
Chris@0 848 void FeatureExtractionManager::finish()
Chris@0 849 {
Chris@0 850 for (PluginMap::iterator pi = m_plugins.begin();
Chris@0 851 pi != m_plugins.end(); ++pi) {
Chris@0 852
Chris@0 853 for (TransformWriterMap::iterator ti = pi->second.begin();
Chris@0 854 ti != pi->second.end(); ++ti) {
Chris@0 855
Chris@0 856 vector<FeatureWriter *> &writers = ti->second;
Chris@0 857
Chris@0 858 for (int i = 0; i < (int)writers.size(); ++i) {
Chris@0 859 writers[i]->flush();
Chris@0 860 writers[i]->finish();
Chris@0 861 }
Chris@0 862 }
Chris@0 863 }
Chris@0 864 }
Chris@0 865
Chris@0 866 void FeatureExtractionManager::print(Transform transform) const
Chris@0 867 {
Chris@0 868 QString qs;
Chris@0 869 QTextStream qts(&qs);
Chris@0 870 transform.toXml(qts);
Chris@0 871 cerr << qs.toStdString() << endl;
Chris@0 872 }