1
|
|
2
|
|
3
|
|
4
|
|
5
|
|
6
|
|
7
|
|
8
|
|
9
|
|
10
|
|
11
|
|
12
|
|
13
|
|
14
|
|
15
|
|
16
|
#include "FeatureExtractionManager.h"
|
17
|
|
18
|
#include <vamp-hostsdk/PluginChannelAdapter.h>
|
19
|
#include <vamp-hostsdk/PluginBufferingAdapter.h>
|
20
|
#include <vamp-hostsdk/PluginInputDomainAdapter.h>
|
21
|
#include <vamp-hostsdk/PluginSummarisingAdapter.h>
|
22
|
#include <vamp-hostsdk/PluginWrapper.h>
|
23
|
#include <vamp-hostsdk/PluginLoader.h>
|
24
|
|
25
|
#include "base/Exceptions.h"
|
26
|
|
27
|
#include <algorithm>
|
28
|
#include <iostream>
|
29
|
|
30
|
using namespace std;
|
31
|
|
32
|
using Vamp::Plugin;
|
33
|
using Vamp::PluginBase;
|
34
|
using Vamp::HostExt::PluginLoader;
|
35
|
using Vamp::HostExt::PluginChannelAdapter;
|
36
|
using Vamp::HostExt::PluginBufferingAdapter;
|
37
|
using Vamp::HostExt::PluginInputDomainAdapter;
|
38
|
using Vamp::HostExt::PluginSummarisingAdapter;
|
39
|
using Vamp::HostExt::PluginWrapper;
|
40
|
|
41
|
#include "data/fileio/FileSource.h"
|
42
|
#include "data/fileio/AudioFileReader.h"
|
43
|
#include "data/fileio/AudioFileReaderFactory.h"
|
44
|
#include "data/fileio/PlaylistFileReader.h"
|
45
|
#include "base/TempDirectory.h"
|
46
|
#include "base/ProgressPrinter.h"
|
47
|
#include "transform/TransformFactory.h"
|
48
|
#include "rdf/RDFTransformFactory.h"
|
49
|
#include "transform/FeatureWriter.h"
|
50
|
|
51
|
#include <QTextStream>
|
52
|
#include <QFile>
|
53
|
#include <QFileInfo>
|
54
|
|
55
|
FeatureExtractionManager::FeatureExtractionManager() :
|
56
|
m_summariesOnly(false),
|
57
|
|
58
|
|
59
|
|
60
|
m_blockSize(16384),
|
61
|
m_defaultSampleRate(0),
|
62
|
m_sampleRate(0),
|
63
|
m_channels(0)
|
64
|
{
|
65
|
}
|
66
|
|
67
|
FeatureExtractionManager::~FeatureExtractionManager()
|
68
|
{
|
69
|
for (PluginMap::iterator pi = m_plugins.begin();
|
70
|
pi != m_plugins.end(); ++pi) {
|
71
|
delete pi->first;
|
72
|
}
|
73
|
foreach (AudioFileReader *r, m_readyReaders) {
|
74
|
delete r;
|
75
|
}
|
76
|
}
|
77
|
|
78
|
void FeatureExtractionManager::setChannels(int channels)
|
79
|
{
|
80
|
m_channels = channels;
|
81
|
}
|
82
|
|
83
|
void FeatureExtractionManager::setDefaultSampleRate(int sampleRate)
|
84
|
{
|
85
|
m_defaultSampleRate = sampleRate;
|
86
|
}
|
87
|
|
88
|
static PluginSummarisingAdapter::SummaryType
|
89
|
getSummaryType(string name)
|
90
|
{
|
91
|
if (name == "min") return PluginSummarisingAdapter::Minimum;
|
92
|
if (name == "max") return PluginSummarisingAdapter::Maximum;
|
93
|
if (name == "mean") return PluginSummarisingAdapter::Mean;
|
94
|
if (name == "median") return PluginSummarisingAdapter::Median;
|
95
|
if (name == "mode") return PluginSummarisingAdapter::Mode;
|
96
|
if (name == "sum") return PluginSummarisingAdapter::Sum;
|
97
|
if (name == "variance") return PluginSummarisingAdapter::Variance;
|
98
|
if (name == "sd") return PluginSummarisingAdapter::StandardDeviation;
|
99
|
if (name == "count") return PluginSummarisingAdapter::Count;
|
100
|
return PluginSummarisingAdapter::UnknownSummaryType;
|
101
|
}
|
102
|
|
103
|
bool FeatureExtractionManager::setSummaryTypes(const set<string> &names,
|
104
|
bool summariesOnly,
|
105
|
const PluginSummarisingAdapter::SegmentBoundaries &boundaries)
|
106
|
{
|
107
|
for (SummaryNameSet::const_iterator i = names.begin();
|
108
|
i != names.end(); ++i) {
|
109
|
if (getSummaryType(*i) == PluginSummarisingAdapter::UnknownSummaryType) {
|
110
|
cerr << "ERROR: Unknown summary type \"" << *i << "\"" << endl;
|
111
|
return false;
|
112
|
}
|
113
|
}
|
114
|
m_summaries = names;
|
115
|
m_summariesOnly = summariesOnly;
|
116
|
m_boundaries = boundaries;
|
117
|
return true;
|
118
|
}
|
119
|
|
120
|
static PluginInputDomainAdapter::WindowType
|
121
|
convertWindowType(WindowType t)
|
122
|
{
|
123
|
switch (t) {
|
124
|
case RectangularWindow:
|
125
|
return PluginInputDomainAdapter::RectangularWindow;
|
126
|
case BartlettWindow:
|
127
|
return PluginInputDomainAdapter::BartlettWindow;
|
128
|
case HammingWindow:
|
129
|
return PluginInputDomainAdapter::HammingWindow;
|
130
|
case HanningWindow:
|
131
|
return PluginInputDomainAdapter::HanningWindow;
|
132
|
case BlackmanWindow:
|
133
|
return PluginInputDomainAdapter::BlackmanWindow;
|
134
|
case NuttallWindow:
|
135
|
return PluginInputDomainAdapter::NuttallWindow;
|
136
|
case BlackmanHarrisWindow:
|
137
|
return PluginInputDomainAdapter::BlackmanHarrisWindow;
|
138
|
default:
|
139
|
cerr << "ERROR: Unknown or unsupported window type \"" << t << "\", using Hann (\"" << HanningWindow << "\")" << endl;
|
140
|
return PluginInputDomainAdapter::HanningWindow;
|
141
|
}
|
142
|
}
|
143
|
|
144
|
bool FeatureExtractionManager::addFeatureExtractor
|
145
|
(Transform transform, const vector<FeatureWriter*> &writers)
|
146
|
{
|
147
|
|
148
|
|
149
|
if (transform.getSampleRate() == 0) {
|
150
|
if (m_sampleRate == 0) {
|
151
|
cerr << "NOTE: Transform does not specify a sample rate, using default rate of " << m_defaultSampleRate << endl;
|
152
|
transform.setSampleRate(m_defaultSampleRate);
|
153
|
m_sampleRate = m_defaultSampleRate;
|
154
|
} else {
|
155
|
cerr << "NOTE: Transform does not specify a sample rate, using previous transform's rate of " << m_sampleRate << endl;
|
156
|
transform.setSampleRate(m_sampleRate);
|
157
|
}
|
158
|
}
|
159
|
|
160
|
if (m_sampleRate == 0) {
|
161
|
m_sampleRate = transform.getSampleRate();
|
162
|
}
|
163
|
|
164
|
if (transform.getSampleRate() != m_sampleRate) {
|
165
|
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;
|
166
|
cerr << "WARNING: Using previous rate of " << m_sampleRate << " for this transform as well" << endl;
|
167
|
transform.setSampleRate(m_sampleRate);
|
168
|
}
|
169
|
|
170
|
Plugin *plugin = 0;
|
171
|
|
172
|
|
173
|
|
174
|
|
175
|
|
176
|
Transform originalTransform = transform;
|
177
|
|
178
|
if (m_transformPluginMap.find(transform) == m_transformPluginMap.end()) {
|
179
|
|
180
|
|
181
|
|
182
|
|
183
|
|
184
|
|
185
|
for (TransformPluginMap::iterator i = m_transformPluginMap.begin();
|
186
|
i != m_transformPluginMap.end(); ++i) {
|
187
|
Transform test = i->first;
|
188
|
test.setOutput(transform.getOutput());
|
189
|
test.setSummaryType(transform.getSummaryType());
|
190
|
if (transform == test) {
|
191
|
cerr << "NOTE: Already have transform identical to this one (for \""
|
192
|
<< transform.getIdentifier().toStdString()
|
193
|
<< "\") in every detail except output identifier and/or "
|
194
|
<< "summary type; sharing its plugin instance" << endl;
|
195
|
plugin = i->second;
|
196
|
if (transform.getSummaryType() != Transform::NoSummary &&
|
197
|
!dynamic_cast<PluginSummarisingAdapter *>(plugin)) {
|
198
|
plugin = new PluginSummarisingAdapter(plugin);
|
199
|
i->second = plugin;
|
200
|
}
|
201
|
break;
|
202
|
}
|
203
|
}
|
204
|
|
205
|
if (!plugin) {
|
206
|
|
207
|
TransformFactory *tf = TransformFactory::getInstance();
|
208
|
|
209
|
PluginBase *pb = tf->instantiatePluginFor(transform);
|
210
|
plugin = tf->downcastVampPlugin(pb);
|
211
|
if (!plugin) {
|
212
|
|
213
|
|
214
|
cerr << "ERROR: Failed to load plugin for transform \""
|
215
|
<< transform.getIdentifier().toStdString() << "\"" << endl;
|
216
|
delete pb;
|
217
|
return false;
|
218
|
}
|
219
|
|
220
|
|
221
|
|
222
|
|
223
|
|
224
|
|
225
|
|
226
|
|
227
|
|
228
|
|
229
|
|
230
|
|
231
|
size_t pluginStepSize = plugin->getPreferredStepSize();
|
232
|
size_t pluginBlockSize = plugin->getPreferredBlockSize();
|
233
|
|
234
|
PluginInputDomainAdapter *pida = 0;
|
235
|
|
236
|
|
237
|
if (plugin->getInputDomain() == Plugin::FrequencyDomain) {
|
238
|
|
239
|
pida = new PluginInputDomainAdapter(plugin);
|
240
|
pida->setProcessTimestampMethod(PluginInputDomainAdapter::ShiftData);
|
241
|
|
242
|
PluginInputDomainAdapter::WindowType wtype =
|
243
|
convertWindowType(transform.getWindowType());
|
244
|
pida->setWindowType(wtype);
|
245
|
plugin = pida;
|
246
|
}
|
247
|
|
248
|
PluginBufferingAdapter *pba = new PluginBufferingAdapter(plugin);
|
249
|
plugin = pba;
|
250
|
|
251
|
if (transform.getStepSize() != 0) {
|
252
|
pba->setPluginStepSize(transform.getStepSize());
|
253
|
} else {
|
254
|
transform.setStepSize(pluginStepSize);
|
255
|
}
|
256
|
|
257
|
if (transform.getBlockSize() != 0) {
|
258
|
pba->setPluginBlockSize(transform.getBlockSize());
|
259
|
} else {
|
260
|
transform.setBlockSize(pluginBlockSize);
|
261
|
}
|
262
|
|
263
|
plugin = new PluginChannelAdapter(plugin);
|
264
|
|
265
|
if (!m_summaries.empty() ||
|
266
|
transform.getSummaryType() != Transform::NoSummary) {
|
267
|
PluginSummarisingAdapter *adapter =
|
268
|
new PluginSummarisingAdapter(plugin);
|
269
|
adapter->setSummarySegmentBoundaries(m_boundaries);
|
270
|
plugin = adapter;
|
271
|
}
|
272
|
|
273
|
if (!plugin->initialise(m_channels, m_blockSize, m_blockSize)) {
|
274
|
cerr << "ERROR: Plugin initialise (channels = " << m_channels << ", stepSize = " << m_blockSize << ", blockSize = " << m_blockSize << ") failed." << endl;
|
275
|
delete plugin;
|
276
|
return false;
|
277
|
}
|
278
|
|
279
|
|
280
|
|
281
|
size_t actualStepSize = 0;
|
282
|
size_t actualBlockSize = 0;
|
283
|
pba->getActualStepAndBlockSizes(actualStepSize, actualBlockSize);
|
284
|
transform.setStepSize(actualStepSize);
|
285
|
transform.setBlockSize(actualBlockSize);
|
286
|
|
287
|
Plugin::OutputList outputs = plugin->getOutputDescriptors();
|
288
|
for (int i = 0; i < (int)outputs.size(); ++i) {
|
289
|
|
290
|
|
291
|
|
292
|
m_pluginOutputs[plugin][outputs[i].identifier] = outputs[i];
|
293
|
m_pluginOutputIndices[outputs[i].identifier] = i;
|
294
|
}
|
295
|
|
296
|
cerr << "NOTE: Loaded and initialised plugin for transform \""
|
297
|
<< transform.getIdentifier().toStdString()
|
298
|
<< "\" with plugin step size " << actualStepSize
|
299
|
<< " and block size " << actualBlockSize
|
300
|
<< " (adapter step and block size " << m_blockSize << ")"
|
301
|
<< endl;
|
302
|
|
303
|
if (pida) {
|
304
|
cerr << "NOTE: PluginInputDomainAdapter timestamp adjustment is "
|
305
|
|
306
|
<< pida->getTimestampAdjustment() << endl;
|
307
|
}
|
308
|
|
309
|
} else {
|
310
|
|
311
|
if (transform.getStepSize() == 0 || transform.getBlockSize() == 0) {
|
312
|
|
313
|
PluginWrapper *pw = dynamic_cast<PluginWrapper *>(plugin);
|
314
|
if (pw) {
|
315
|
PluginBufferingAdapter *pba =
|
316
|
pw->getWrapper<PluginBufferingAdapter>();
|
317
|
if (pba) {
|
318
|
size_t actualStepSize = 0;
|
319
|
size_t actualBlockSize = 0;
|
320
|
pba->getActualStepAndBlockSizes(actualStepSize,
|
321
|
actualBlockSize);
|
322
|
if (transform.getStepSize() == 0) {
|
323
|
transform.setStepSize(actualStepSize);
|
324
|
}
|
325
|
if (transform.getBlockSize() == 0) {
|
326
|
transform.setBlockSize(actualBlockSize);
|
327
|
}
|
328
|
}
|
329
|
}
|
330
|
}
|
331
|
}
|
332
|
|
333
|
if (transform.getOutput() == "") {
|
334
|
transform.setOutput
|
335
|
(plugin->getOutputDescriptors()[0].identifier.c_str());
|
336
|
}
|
337
|
|
338
|
m_transformPluginMap[transform] = plugin;
|
339
|
|
340
|
if (!(originalTransform == transform)) {
|
341
|
m_transformPluginMap[originalTransform] = plugin;
|
342
|
}
|
343
|
|
344
|
} else {
|
345
|
|
346
|
plugin = m_transformPluginMap[transform];
|
347
|
}
|
348
|
|
349
|
m_plugins[plugin][transform] = writers;
|
350
|
|
351
|
return true;
|
352
|
}
|
353
|
|
354
|
bool FeatureExtractionManager::addDefaultFeatureExtractor
|
355
|
(TransformId transformId, const vector<FeatureWriter*> &writers)
|
356
|
{
|
357
|
TransformFactory *tf = TransformFactory::getInstance();
|
358
|
|
359
|
if (m_sampleRate == 0) {
|
360
|
if (m_defaultSampleRate == 0) {
|
361
|
cerr << "ERROR: Default transform requested, but no default sample rate available" << endl;
|
362
|
return false;
|
363
|
} else {
|
364
|
cerr << "NOTE: Using default sample rate of " << m_defaultSampleRate << " for default transform" << endl;
|
365
|
m_sampleRate = m_defaultSampleRate;
|
366
|
}
|
367
|
}
|
368
|
|
369
|
Transform transform = tf->getDefaultTransformFor(transformId, m_sampleRate);
|
370
|
|
371
|
return addFeatureExtractor(transform, writers);
|
372
|
}
|
373
|
|
374
|
|
375
|
bool FeatureExtractionManager::addFeatureExtractorFromFile
|
376
|
(QString transformXmlFile, const vector<FeatureWriter*> &writers)
|
377
|
{
|
378
|
RDFTransformFactory factory
|
379
|
(QUrl::fromLocalFile(QFileInfo(transformXmlFile).absoluteFilePath())
|
380
|
.toString());
|
381
|
ProgressPrinter printer("Parsing transforms RDF file");
|
382
|
std::vector<Transform> transforms = factory.getTransforms(&printer);
|
383
|
if (!factory.isOK()) {
|
384
|
cerr << "WARNING: FeatureExtractionManager::addFeatureExtractorFromFile: Failed to parse transforms file: " << factory.getErrorString().toStdString() << endl;
|
385
|
if (factory.isRDF()) {
|
386
|
return false;
|
387
|
}
|
388
|
}
|
389
|
if (!transforms.empty()) {
|
390
|
std::sort(transforms.begin(), transforms.end());
|
391
|
bool success = true;
|
392
|
for (int i = 0; i < (int)transforms.size(); ++i) {
|
393
|
if (!addFeatureExtractor(transforms[i], writers)) {
|
394
|
success = false;
|
395
|
}
|
396
|
}
|
397
|
cerr << "\n-------------------\nTransform order:\n";
|
398
|
for (int i = 0; i < (int)transforms.size(); ++i) {
|
399
|
cerr << i << "\t" << transforms[i].getIdentifier() << "\n";
|
400
|
}
|
401
|
cerr << "-------------------\n";
|
402
|
return success;
|
403
|
}
|
404
|
|
405
|
QFile file(transformXmlFile);
|
406
|
if (!file.open(QIODevice::ReadOnly | QIODevice::Text)) {
|
407
|
cerr << "ERROR: Failed to open transform XML file \""
|
408
|
<< transformXmlFile.toStdString() << "\" for reading" << endl;
|
409
|
return false;
|
410
|
}
|
411
|
|
412
|
QTextStream *qts = new QTextStream(&file);
|
413
|
QString qs = qts->readAll();
|
414
|
delete qts;
|
415
|
file.close();
|
416
|
|
417
|
Transform transform(qs);
|
418
|
|
419
|
return addFeatureExtractor(transform, writers);
|
420
|
}
|
421
|
|
422
|
void FeatureExtractionManager::addSource(QString audioSource)
|
423
|
{
|
424
|
if (QFileInfo(audioSource).suffix().toLower() == "m3u") {
|
425
|
ProgressPrinter retrievalProgress("Opening playlist file...");
|
426
|
FileSource source(audioSource, &retrievalProgress);
|
427
|
if (!source.isAvailable()) {
|
428
|
cerr << "ERROR: File or URL \"" << audioSource.toStdString()
|
429
|
<< "\" could not be located" << endl;
|
430
|
throw FileNotFound(audioSource);
|
431
|
}
|
432
|
source.waitForData();
|
433
|
PlaylistFileReader reader(source);
|
434
|
if (reader.isOK()) {
|
435
|
vector<QString> files = reader.load();
|
436
|
for (int i = 0; i < (int)files.size(); ++i) {
|
437
|
addSource(files[i]);
|
438
|
}
|
439
|
return;
|
440
|
} else {
|
441
|
cerr << "ERROR: Playlist \"" << audioSource.toStdString()
|
442
|
<< "\" could not be opened" << endl;
|
443
|
throw FileNotFound(audioSource);
|
444
|
}
|
445
|
}
|
446
|
|
447
|
std::cerr << "Have audio source: \"" << audioSource.toStdString() << "\"" << std::endl;
|
448
|
|
449
|
|
450
|
|
451
|
|
452
|
|
453
|
if (m_channels == 0 || m_defaultSampleRate == 0) {
|
454
|
|
455
|
ProgressPrinter retrievalProgress("Determining default rate and channel count from first input file...");
|
456
|
|
457
|
FileSource source(audioSource, &retrievalProgress);
|
458
|
if (!source.isAvailable()) {
|
459
|
cerr << "ERROR: File or URL \"" << audioSource.toStdString()
|
460
|
<< "\" could not be located" << endl;
|
461
|
throw FileNotFound(audioSource);
|
462
|
}
|
463
|
|
464
|
source.waitForData();
|
465
|
|
466
|
|
467
|
|
468
|
|
469
|
AudioFileReader *reader =
|
470
|
AudioFileReaderFactory::createReader(source, 0, &retrievalProgress);
|
471
|
|
472
|
if (!reader) {
|
473
|
throw FailedToOpenFile(audioSource);
|
474
|
}
|
475
|
|
476
|
retrievalProgress.done();
|
477
|
|
478
|
cerr << "File or URL \"" << audioSource.toStdString() << "\" opened successfully" << endl;
|
479
|
|
480
|
if (m_channels == 0) {
|
481
|
m_channels = reader->getChannelCount();
|
482
|
cerr << "Taking default channel count of "
|
483
|
<< reader->getChannelCount() << " from file" << endl;
|
484
|
}
|
485
|
|
486
|
if (m_defaultSampleRate == 0) {
|
487
|
m_defaultSampleRate = reader->getNativeRate();
|
488
|
cerr << "Taking default sample rate of "
|
489
|
<< reader->getNativeRate() << "Hz from file" << endl;
|
490
|
cerr << "(Note: Default may be overridden by transforms)" << endl;
|
491
|
}
|
492
|
|
493
|
m_readyReaders[audioSource] = reader;
|
494
|
}
|
495
|
}
|
496
|
|
497
|
void FeatureExtractionManager::extractFeatures(QString audioSource, bool force)
|
498
|
{
|
499
|
if (m_plugins.empty()) return;
|
500
|
|
501
|
if (QFileInfo(audioSource).suffix().toLower() == "m3u") {
|
502
|
FileSource source(audioSource);
|
503
|
PlaylistFileReader reader(source);
|
504
|
if (reader.isOK()) {
|
505
|
vector<QString> files = reader.load();
|
506
|
for (int i = 0; i < (int)files.size(); ++i) {
|
507
|
try {
|
508
|
extractFeatures(files[i], force);
|
509
|
} catch (const std::exception &e) {
|
510
|
if (!force) throw;
|
511
|
cerr << "ERROR: Feature extraction failed for playlist entry \""
|
512
|
<< files[i].toStdString()
|
513
|
<< "\": " << e.what() << endl;
|
514
|
|
515
|
if (++i != files.size()) {
|
516
|
cerr << "NOTE: \"--force\" option was provided, continuing (more errors may occur)" << endl;
|
517
|
}
|
518
|
}
|
519
|
}
|
520
|
return;
|
521
|
} else {
|
522
|
cerr << "ERROR: Playlist \"" << audioSource.toStdString()
|
523
|
<< "\" could not be opened" << endl;
|
524
|
throw FileNotFound(audioSource);
|
525
|
}
|
526
|
}
|
527
|
|
528
|
testOutputFiles(audioSource);
|
529
|
|
530
|
if (m_sampleRate == 0) {
|
531
|
throw FileOperationFailed
|
532
|
(audioSource, "internal error: have sources and plugins, but no sample rate");
|
533
|
}
|
534
|
if (m_channels == 0) {
|
535
|
throw FileOperationFailed
|
536
|
(audioSource, "internal error: have sources and plugins, but no channel count");
|
537
|
}
|
538
|
|
539
|
AudioFileReader *reader = 0;
|
540
|
|
541
|
if (m_readyReaders.contains(audioSource)) {
|
542
|
reader = m_readyReaders[audioSource];
|
543
|
m_readyReaders.remove(audioSource);
|
544
|
if (reader->getChannelCount() != m_channels ||
|
545
|
reader->getSampleRate() != m_sampleRate) {
|
546
|
|
547
|
delete reader;
|
548
|
reader = 0;
|
549
|
}
|
550
|
}
|
551
|
if (!reader) {
|
552
|
ProgressPrinter retrievalProgress("Retrieving audio data...");
|
553
|
FileSource source(audioSource, &retrievalProgress);
|
554
|
source.waitForData();
|
555
|
reader = AudioFileReaderFactory::createReader
|
556
|
(source, m_sampleRate, &retrievalProgress);
|
557
|
retrievalProgress.done();
|
558
|
}
|
559
|
|
560
|
if (!reader) {
|
561
|
throw FailedToOpenFile(audioSource);
|
562
|
}
|
563
|
|
564
|
cerr << "Audio file \"" << audioSource.toStdString() << "\": "
|
565
|
<< reader->getChannelCount() << "ch at "
|
566
|
<< reader->getNativeRate() << "Hz" << endl;
|
567
|
if (reader->getChannelCount() != m_channels ||
|
568
|
reader->getNativeRate() != m_sampleRate) {
|
569
|
cerr << "NOTE: File will be mixed or resampled for processing: "
|
570
|
<< m_channels << "ch at "
|
571
|
<< m_sampleRate << "Hz" << endl;
|
572
|
}
|
573
|
|
574
|
|
575
|
float **data = new float *[m_channels];
|
576
|
for (int c = 0; c < m_channels; ++c) {
|
577
|
data[c] = new float[m_blockSize];
|
578
|
}
|
579
|
|
580
|
struct LifespanMgr {
|
581
|
|
582
|
AudioFileReader *m_r;
|
583
|
int m_c;
|
584
|
float **m_d;
|
585
|
LifespanMgr(AudioFileReader *r, int c, float **d) :
|
586
|
m_r(r), m_c(c), m_d(d) { }
|
587
|
~LifespanMgr() { destroy(); }
|
588
|
void destroy() {
|
589
|
if (!m_r) return;
|
590
|
delete m_r;
|
591
|
for (int i = 0; i < m_c; ++i) delete[] m_d[i];
|
592
|
delete[] m_d;
|
593
|
m_r = 0;
|
594
|
}
|
595
|
};
|
596
|
LifespanMgr lifemgr(reader, m_channels, data);
|
597
|
|
598
|
size_t frameCount = reader->getFrameCount();
|
599
|
|
600
|
|
601
|
|
602
|
for (PluginMap::iterator pi = m_plugins.begin();
|
603
|
pi != m_plugins.end(); ++pi) {
|
604
|
|
605
|
Plugin *plugin = pi->first;
|
606
|
|
607
|
|
608
|
plugin->reset();
|
609
|
|
610
|
for (TransformWriterMap::iterator ti = pi->second.begin();
|
611
|
ti != pi->second.end(); ++ti) {
|
612
|
|
613
|
const Transform &transform = ti->first;
|
614
|
|
615
|
|
616
|
|
617
|
|
618
|
|
619
|
|
620
|
|
621
|
|
622
|
string outputId = transform.getOutput().toStdString();
|
623
|
if (m_pluginOutputs[plugin].find(outputId) ==
|
624
|
m_pluginOutputs[plugin].end()) {
|
625
|
|
626
|
cerr << "WARNING: Nonexistent plugin output \"" << outputId << "\" requested for transform \""
|
627
|
<< transform.getIdentifier().toStdString() << "\", ignoring this transform"
|
628
|
<< endl;
|
629
|
|
630
|
|
631
|
|
632
|
|
633
|
|
634
|
|
635
|
|
636
|
|
637
|
|
638
|
|
639
|
|
640
|
|
641
|
|
642
|
|
643
|
|
644
|
}
|
645
|
}
|
646
|
}
|
647
|
|
648
|
long startFrame = 0;
|
649
|
long endFrame = frameCount;
|
650
|
|
651
|
|
652
|
|
653
|
|
654
|
|
655
|
|
656
|
|
657
|
|
658
|
|
659
|
|
660
|
|
661
|
|
662
|
|
663
|
|
664
|
|
665
|
|
666
|
|
667
|
|
668
|
|
669
|
|
670
|
|
671
|
|
672
|
|
673
|
|
674
|
|
675
|
|
676
|
for (PluginMap::iterator pi = m_plugins.begin();
|
677
|
pi != m_plugins.end(); ++pi) {
|
678
|
|
679
|
for (TransformWriterMap::const_iterator ti = pi->second.begin();
|
680
|
ti != pi->second.end(); ++ti) {
|
681
|
|
682
|
const vector<FeatureWriter *> &writers = ti->second;
|
683
|
|
684
|
for (int j = 0; j < (int)writers.size(); ++j) {
|
685
|
FeatureWriter::TrackMetadata m;
|
686
|
m.title = reader->getTitle();
|
687
|
m.maker = reader->getMaker();
|
688
|
if (m.title != "" && m.maker != "") {
|
689
|
writers[j]->setTrackMetadata(audioSource, m);
|
690
|
}
|
691
|
}
|
692
|
}
|
693
|
}
|
694
|
|
695
|
ProgressPrinter extractionProgress("Extracting and writing features...");
|
696
|
int progress = 0;
|
697
|
|
698
|
for (long i = startFrame; i < endFrame; i += m_blockSize) {
|
699
|
|
700
|
|
701
|
|
702
|
|
703
|
SampleBlock frames;
|
704
|
reader->getInterleavedFrames(i, m_blockSize, frames);
|
705
|
|
706
|
|
707
|
|
708
|
|
709
|
|
710
|
|
711
|
|
712
|
int rc = reader->getChannelCount();
|
713
|
|
714
|
|
715
|
|
716
|
int index;
|
717
|
int fc = (int)frames.size();
|
718
|
|
719
|
if (m_channels == 1) {
|
720
|
for (int j = 0; j < m_blockSize; ++j) {
|
721
|
data[0][j] = 0.f;
|
722
|
}
|
723
|
for (int c = 0; c < rc; ++c) {
|
724
|
for (int j = 0; j < m_blockSize; ++j) {
|
725
|
index = j * rc + c;
|
726
|
if (index < fc) data[0][j] += frames[index];
|
727
|
}
|
728
|
}
|
729
|
for (int j = 0; j < m_blockSize; ++j) {
|
730
|
data[0][j] /= rc;
|
731
|
}
|
732
|
} else {
|
733
|
for (int c = 0; c < m_channels; ++c) {
|
734
|
for (int j = 0; j < m_blockSize; ++j) {
|
735
|
data[c][j] = 0.f;
|
736
|
}
|
737
|
if (c < rc) {
|
738
|
for (int j = 0; j < m_blockSize; ++j) {
|
739
|
index = j * rc + c;
|
740
|
if (index < fc) data[c][j] += frames[index];
|
741
|
}
|
742
|
}
|
743
|
}
|
744
|
}
|
745
|
|
746
|
Vamp::RealTime timestamp = Vamp::RealTime::frame2RealTime
|
747
|
(i, m_sampleRate);
|
748
|
|
749
|
for (PluginMap::iterator pi = m_plugins.begin();
|
750
|
pi != m_plugins.end(); ++pi) {
|
751
|
|
752
|
Plugin *plugin = pi->first;
|
753
|
Plugin::FeatureSet featureSet = plugin->process(data, timestamp);
|
754
|
|
755
|
if (!m_summariesOnly) {
|
756
|
writeFeatures(audioSource, plugin, featureSet);
|
757
|
}
|
758
|
}
|
759
|
|
760
|
int pp = progress;
|
761
|
progress = int(((i - startFrame) * 100.0) / (endFrame - startFrame) + 0.1);
|
762
|
if (progress > pp) extractionProgress.setProgress(progress);
|
763
|
}
|
764
|
|
765
|
|
766
|
|
767
|
lifemgr.destroy();
|
768
|
|
769
|
|
770
|
|
771
|
|
772
|
|
773
|
|
774
|
|
775
|
map<Transform, pair<Vamp::Plugin *, TransformWriterMap> > tempMap;
|
776
|
|
777
|
for (PluginMap::iterator pi = m_plugins.begin();
|
778
|
pi != m_plugins.end(); ++pi) {
|
779
|
pair<Vamp::Plugin *, TransformWriterMap> tempTransformWriterPair = (*pi);
|
780
|
Transform tempTransform = (pi->second).begin()->first;
|
781
|
pair<Transform, pair<Vamp::Plugin *, TransformWriterMap> > tempPair (tempTransform,tempTransformWriterPair);
|
782
|
tempMap.insert(tempPair);
|
783
|
}
|
784
|
|
785
|
for (map<Transform, pair<Vamp::Plugin *, TransformWriterMap> >::iterator superPi = tempMap.begin();
|
786
|
superPi != tempMap.end(); ++superPi) {
|
787
|
|
788
|
pair<Vamp::Plugin *, TransformWriterMap> pi = superPi->second;
|
789
|
|
790
|
|
791
|
|
792
|
Plugin *plugin = pi.first;
|
793
|
Plugin::FeatureSet featureSet = plugin->getRemainingFeatures();
|
794
|
|
795
|
if (!m_summariesOnly) {
|
796
|
writeFeatures(audioSource, plugin, featureSet);
|
797
|
}
|
798
|
|
799
|
if (!m_summaries.empty()) {
|
800
|
PluginSummarisingAdapter *adapter =
|
801
|
dynamic_cast<PluginSummarisingAdapter *>(plugin);
|
802
|
if (!adapter) {
|
803
|
cerr << "WARNING: Summaries requested, but plugin is not a summarising adapter" << endl;
|
804
|
} else {
|
805
|
for (SummaryNameSet::const_iterator sni = m_summaries.begin();
|
806
|
sni != m_summaries.end(); ++sni) {
|
807
|
featureSet.clear();
|
808
|
|
809
|
|
810
|
|
811
|
|
812
|
|
813
|
featureSet = adapter->getSummaryForAllOutputs
|
814
|
(getSummaryType(*sni),
|
815
|
PluginSummarisingAdapter::ContinuousTimeAverage);
|
816
|
writeFeatures(audioSource, plugin, featureSet,
|
817
|
Transform::stringToSummaryType(sni->c_str()));
|
818
|
}
|
819
|
}
|
820
|
}
|
821
|
|
822
|
writeSummaries(audioSource, plugin);
|
823
|
}
|
824
|
|
825
|
extractionProgress.done();
|
826
|
|
827
|
finish();
|
828
|
|
829
|
TempDirectory::getInstance()->cleanup();
|
830
|
}
|
831
|
|
832
|
void
|
833
|
FeatureExtractionManager::writeSummaries(QString audioSource, Plugin *plugin)
|
834
|
{
|
835
|
|
836
|
PluginMap::iterator pi = m_plugins.find(plugin);
|
837
|
|
838
|
for (TransformWriterMap::const_iterator ti = pi->second.begin();
|
839
|
ti != pi->second.end(); ++ti) {
|
840
|
|
841
|
const Transform &transform = ti->first;
|
842
|
const vector<FeatureWriter *> &writers = ti->second;
|
843
|
|
844
|
Transform::SummaryType summaryType = transform.getSummaryType();
|
845
|
PluginSummarisingAdapter::SummaryType pType =
|
846
|
(PluginSummarisingAdapter::SummaryType)summaryType;
|
847
|
|
848
|
if (transform.getSummaryType() == Transform::NoSummary) {
|
849
|
continue;
|
850
|
}
|
851
|
|
852
|
PluginSummarisingAdapter *adapter =
|
853
|
dynamic_cast<PluginSummarisingAdapter *>(plugin);
|
854
|
if (!adapter) {
|
855
|
cerr << "FeatureExtractionManager::writeSummaries: INTERNAL ERROR: Summary requested for transform, but plugin is not a summarising adapter" << endl;
|
856
|
continue;
|
857
|
}
|
858
|
|
859
|
Plugin::FeatureSet featureSet = adapter->getSummaryForAllOutputs
|
860
|
(pType, PluginSummarisingAdapter::ContinuousTimeAverage);
|
861
|
|
862
|
|
863
|
|
864
|
writeFeatures(audioSource, plugin, featureSet, summaryType);
|
865
|
}
|
866
|
}
|
867
|
|
868
|
void FeatureExtractionManager::writeFeatures(QString audioSource,
|
869
|
Plugin *plugin,
|
870
|
const Plugin::FeatureSet &features,
|
871
|
Transform::SummaryType summaryType)
|
872
|
{
|
873
|
|
874
|
PluginMap::iterator pi = m_plugins.find(plugin);
|
875
|
|
876
|
for (TransformWriterMap::const_iterator ti = pi->second.begin();
|
877
|
ti != pi->second.end(); ++ti) {
|
878
|
|
879
|
const Transform &transform = ti->first;
|
880
|
const vector<FeatureWriter *> &writers = ti->second;
|
881
|
|
882
|
if (transform.getSummaryType() != Transform::NoSummary &&
|
883
|
m_summaries.empty() &&
|
884
|
summaryType == Transform::NoSummary) {
|
885
|
continue;
|
886
|
}
|
887
|
|
888
|
if (transform.getSummaryType() != Transform::NoSummary &&
|
889
|
summaryType != Transform::NoSummary &&
|
890
|
transform.getSummaryType() != summaryType) {
|
891
|
continue;
|
892
|
}
|
893
|
|
894
|
string outputId = transform.getOutput().toStdString();
|
895
|
|
896
|
if (m_pluginOutputs[plugin].find(outputId) ==
|
897
|
m_pluginOutputs[plugin].end()) {
|
898
|
continue;
|
899
|
}
|
900
|
|
901
|
const Plugin::OutputDescriptor &desc =
|
902
|
m_pluginOutputs[plugin][outputId];
|
903
|
|
904
|
int outputIndex = m_pluginOutputIndices[outputId];
|
905
|
Plugin::FeatureSet::const_iterator fsi = features.find(outputIndex);
|
906
|
if (fsi == features.end()) continue;
|
907
|
|
908
|
for (int j = 0; j < (int)writers.size(); ++j) {
|
909
|
writers[j]->write
|
910
|
(audioSource, transform, desc, fsi->second,
|
911
|
Transform::summaryTypeToString(summaryType).toStdString());
|
912
|
}
|
913
|
}
|
914
|
}
|
915
|
|
916
|
void FeatureExtractionManager::testOutputFiles(QString audioSource)
|
917
|
{
|
918
|
for (PluginMap::iterator pi = m_plugins.begin();
|
919
|
pi != m_plugins.end(); ++pi) {
|
920
|
|
921
|
for (TransformWriterMap::iterator ti = pi->second.begin();
|
922
|
ti != pi->second.end(); ++ti) {
|
923
|
|
924
|
vector<FeatureWriter *> &writers = ti->second;
|
925
|
|
926
|
for (int i = 0; i < (int)writers.size(); ++i) {
|
927
|
writers[i]->testOutputFile(audioSource, ti->first.getIdentifier());
|
928
|
}
|
929
|
}
|
930
|
}
|
931
|
}
|
932
|
|
933
|
void FeatureExtractionManager::finish()
|
934
|
{
|
935
|
for (PluginMap::iterator pi = m_plugins.begin();
|
936
|
pi != m_plugins.end(); ++pi) {
|
937
|
|
938
|
for (TransformWriterMap::iterator ti = pi->second.begin();
|
939
|
ti != pi->second.end(); ++ti) {
|
940
|
|
941
|
vector<FeatureWriter *> &writers = ti->second;
|
942
|
|
943
|
for (int i = 0; i < (int)writers.size(); ++i) {
|
944
|
writers[i]->flush();
|
945
|
writers[i]->finish();
|
946
|
}
|
947
|
}
|
948
|
}
|
949
|
}
|
950
|
|
951
|
void FeatureExtractionManager::print(Transform transform) const
|
952
|
{
|
953
|
QString qs;
|
954
|
QTextStream qts(&qs);
|
955
|
transform.toXml(qts);
|
956
|
cerr << qs.toStdString() << endl;
|
957
|
}
|