annotate src/MatchVampPlugin.cpp @ 177:937951e66c5b refactors

Add parameters for min & max frequency
author Chris Cannam
date Fri, 13 Feb 2015 15:27:34 +0000
parents e0dc4675059c
children d1bc89794cd4
rev   line source
cannam@0 1 /* -*- c-basic-offset: 4 indent-tabs-mode: nil -*- vi:set ts=8 sts=4 sw=4: */
cannam@0 2
cannam@0 3 /*
cannam@0 4 Vamp feature extraction plugin using the MATCH audio alignment
cannam@0 5 algorithm.
cannam@0 6
cannam@0 7 Centre for Digital Music, Queen Mary, University of London.
cannam@0 8 This file copyright 2007 Simon Dixon, Chris Cannam and QMUL.
cannam@0 9
cannam@0 10 This program is free software; you can redistribute it and/or
cannam@0 11 modify it under the terms of the GNU General Public License as
cannam@0 12 published by the Free Software Foundation; either version 2 of the
cannam@0 13 License, or (at your option) any later version. See the file
cannam@0 14 COPYING included with this distribution for more information.
cannam@0 15 */
cannam@0 16
cannam@0 17 #include "MatchVampPlugin.h"
cannam@0 18
cannam@0 19 #include "Matcher.h"
Chris@74 20 #include "MatchFeatureFeeder.h"
Chris@74 21 #include "FeatureExtractor.h"
cannam@0 22 #include "Path.h"
cannam@0 23
cannam@0 24 #include <vamp/vamp.h>
cannam@0 25 #include <vamp-sdk/PluginAdapter.h>
cannam@0 26 #include <vamp-sdk/RealTime.h>
cannam@0 27
cannam@0 28 #include <vector>
cannam@0 29 #include <algorithm>
cannam@0 30
cannam@0 31 //static int extant = 0;
cannam@0 32
cannam@0 33 #ifdef _WIN32
cannam@0 34 HANDLE
cannam@0 35 MatchVampPlugin::m_serialisingMutex;
cannam@0 36 #else
cannam@0 37 pthread_mutex_t
cannam@0 38 MatchVampPlugin::m_serialisingMutex;
cannam@0 39 #endif
cannam@0 40
cannam@0 41 bool
cannam@0 42 MatchVampPlugin::m_serialisingMutexInitialised = false;
cannam@0 43
Chris@10 44 // We want to ensure our freq map / crossover bin in Matcher.cpp are
Chris@10 45 // always valid with a fixed FFT length in seconds, so must reject low
Chris@10 46 // sample rates
Chris@10 47 static float sampleRateMin = 5000.f;
Chris@10 48
Chris@52 49 static float defaultStepTime = 0.020f;
Chris@15 50
cannam@0 51 MatchVampPlugin::MatchVampPlugin(float inputSampleRate) :
cannam@0 52 Plugin(inputSampleRate),
Chris@52 53 m_stepSize(int(inputSampleRate * defaultStepTime + 0.001)),
Chris@15 54 m_stepTime(defaultStepTime),
Chris@16 55 m_blockSize(2048),
cannam@0 56 m_serialise(false),
cannam@0 57 m_begin(true),
Chris@17 58 m_locked(false),
Chris@138 59 m_smooth(false),
Chris@74 60 m_frameNo(0),
Chris@113 61 m_params(defaultStepTime),
Chris@113 62 m_defaultParams(defaultStepTime),
Chris@38 63 m_feParams(inputSampleRate, m_blockSize),
Chris@103 64 m_defaultFeParams(inputSampleRate, m_blockSize),
Chris@161 65 m_secondReferenceFrequency(m_defaultFeParams.referenceFrequency),
Chris@103 66 m_fcParams(),
Chris@143 67 m_defaultFcParams(),
Chris@143 68 m_dParams(),
Chris@143 69 m_defaultDParams()
cannam@0 70 {
Chris@10 71 if (inputSampleRate < sampleRateMin) {
Chris@10 72 std::cerr << "MatchVampPlugin::MatchVampPlugin: input sample rate "
Chris@10 73 << inputSampleRate << " < min supported rate "
Chris@10 74 << sampleRateMin << ", plugin will refuse to initialise"
Chris@10 75 << std::endl;
Chris@10 76 }
Chris@10 77
cannam@0 78 if (!m_serialisingMutexInitialised) {
cannam@0 79 m_serialisingMutexInitialised = true;
cannam@0 80 #ifdef _WIN32
cannam@0 81 m_serialisingMutex = CreateMutex(NULL, FALSE, NULL);
cannam@0 82 #else
cannam@0 83 pthread_mutex_init(&m_serialisingMutex, 0);
cannam@0 84 #endif
cannam@0 85 }
cannam@0 86
Chris@107 87 m_pipeline = 0;
cannam@0 88 // std::cerr << "MatchVampPlugin::MatchVampPlugin(" << this << "): extant = " << ++extant << std::endl;
cannam@0 89 }
cannam@0 90
cannam@0 91 MatchVampPlugin::~MatchVampPlugin()
cannam@0 92 {
cannam@0 93 // std::cerr << "MatchVampPlugin::~MatchVampPlugin(" << this << "): extant = " << --extant << std::endl;
cannam@0 94
Chris@107 95 delete m_pipeline;
cannam@0 96
cannam@0 97 if (m_locked) {
cannam@0 98 #ifdef _WIN32
cannam@0 99 ReleaseMutex(m_serialisingMutex);
cannam@0 100 #else
cannam@0 101 pthread_mutex_unlock(&m_serialisingMutex);
cannam@0 102 #endif
cannam@0 103 m_locked = false;
cannam@0 104 }
cannam@0 105 }
cannam@0 106
cannam@0 107 string
cannam@0 108 MatchVampPlugin::getIdentifier() const
cannam@0 109 {
cannam@0 110 return "match";
cannam@0 111 }
cannam@0 112
cannam@0 113 string
cannam@0 114 MatchVampPlugin::getName() const
cannam@0 115 {
cannam@0 116 return "Match Performance Aligner";
cannam@0 117 }
cannam@0 118
cannam@0 119 string
cannam@0 120 MatchVampPlugin::getDescription() const
cannam@0 121 {
cannam@0 122 return "Calculate alignment between two performances in separate channel inputs";
cannam@0 123 }
cannam@0 124
cannam@0 125 string
cannam@0 126 MatchVampPlugin::getMaker() const
cannam@0 127 {
cannam@0 128 return "Simon Dixon (plugin by Chris Cannam)";
cannam@0 129 }
cannam@0 130
cannam@0 131 int
cannam@0 132 MatchVampPlugin::getPluginVersion() const
cannam@0 133 {
Chris@137 134 return 3;
cannam@0 135 }
cannam@0 136
cannam@0 137 string
cannam@0 138 MatchVampPlugin::getCopyright() const
cannam@0 139 {
cannam@0 140 return "GPL";
cannam@0 141 }
cannam@0 142
cannam@0 143 MatchVampPlugin::ParameterList
cannam@0 144 MatchVampPlugin::getParameterDescriptors() const
cannam@0 145 {
cannam@0 146 ParameterList list;
cannam@0 147
cannam@0 148 ParameterDescriptor desc;
Chris@18 149
Chris@162 150 desc.identifier = "freq1";
Chris@162 151 desc.name = "Tuning frequency of first input";
Chris@162 152 desc.description = "Tuning frequency (concert A) for the reference audio.";
Chris@162 153 desc.minValue = 220.0;
Chris@162 154 desc.maxValue = 880.0;
Chris@162 155 desc.defaultValue = (float)m_defaultFeParams.referenceFrequency;
Chris@162 156 desc.isQuantized = false;
Chris@162 157 desc.unit = "Hz";
Chris@162 158 list.push_back(desc);
Chris@162 159
Chris@162 160 desc.identifier = "freq2";
Chris@162 161 desc.name = "Tuning frequency of second input";
Chris@162 162 desc.description = "Tuning frequency (concert A) for the other audio.";
Chris@162 163 desc.minValue = 220.0;
Chris@162 164 desc.maxValue = 880.0;
Chris@162 165 desc.defaultValue = (float)m_defaultFeParams.referenceFrequency;
Chris@162 166 desc.isQuantized = false;
Chris@162 167 desc.unit = "Hz";
Chris@162 168 list.push_back(desc);
Chris@162 169
Chris@177 170 desc.identifier = "minfreq";
Chris@177 171 desc.name = "Minimum frequency";
Chris@177 172 desc.description = "Minimum frequency to include in features.";
Chris@177 173 desc.minValue = 0.0;
Chris@177 174 desc.maxValue = (float)m_inputSampleRate / 4.f;
Chris@177 175 desc.defaultValue = (float)m_defaultFeParams.minFrequency;
Chris@177 176 desc.isQuantized = false;
Chris@177 177 desc.unit = "Hz";
Chris@177 178 list.push_back(desc);
Chris@177 179
Chris@177 180 desc.identifier = "maxfreq";
Chris@177 181 desc.name = "Maximum frequency";
Chris@177 182 desc.description = "Maximum frequency to include in features.";
Chris@177 183 desc.minValue = 1000.0;
Chris@177 184 desc.maxValue = (float)m_inputSampleRate / 2.f;
Chris@177 185 desc.defaultValue = (float)m_defaultFeParams.maxFrequency;
Chris@177 186 desc.isQuantized = false;
Chris@177 187 desc.unit = "Hz";
Chris@177 188 list.push_back(desc);
Chris@177 189
Chris@162 190 desc.unit = "";
Chris@162 191
Chris@162 192 desc.identifier = "usechroma";
Chris@162 193 desc.name = "Feature type";
Chris@162 194 desc.description = "Whether to use warped spectrogram or chroma frequency map";
cannam@0 195 desc.minValue = 0;
cannam@0 196 desc.maxValue = 1;
Chris@162 197 desc.defaultValue = m_defaultFeParams.useChromaFrequencyMap ? 1 : 0;
Chris@162 198 desc.isQuantized = true;
Chris@162 199 desc.quantizeStep = 1;
Chris@162 200 desc.valueNames.clear();
Chris@162 201 desc.valueNames.push_back("Spectral");
Chris@162 202 desc.valueNames.push_back("Chroma");
Chris@162 203 list.push_back(desc);
Chris@162 204
Chris@162 205 desc.valueNames.clear();
Chris@162 206
Chris@162 207 desc.identifier = "usespecdiff";
Chris@162 208 desc.name = "Use feature difference";
Chris@162 209 desc.description = "Whether to use half-wave rectified feature-to-feature difference instead of straight spectral or chroma feature";
Chris@162 210 desc.minValue = 0;
Chris@162 211 desc.maxValue = 1;
Chris@162 212 desc.defaultValue = (int)m_defaultFcParams.order;
cannam@0 213 desc.isQuantized = true;
cannam@0 214 desc.quantizeStep = 1;
cannam@0 215 list.push_back(desc);
cannam@0 216
Chris@18 217 desc.identifier = "framenorm";
Chris@162 218 desc.name = "Frame normalisation";
Chris@162 219 desc.description = "Type of normalisation to use for features";
Chris@18 220 desc.minValue = 0;
Chris@18 221 desc.maxValue = 2;
Chris@103 222 desc.defaultValue = (int)m_defaultFcParams.norm;
Chris@18 223 desc.isQuantized = true;
Chris@18 224 desc.quantizeStep = 1;
Chris@18 225 desc.valueNames.clear();
Chris@18 226 desc.valueNames.push_back("None");
Chris@162 227 desc.valueNames.push_back("Sum to 1");
Chris@162 228 desc.valueNames.push_back("Long-term average");
Chris@18 229 list.push_back(desc);
Chris@18 230 desc.valueNames.clear();
Chris@148 231 desc.defaultValue = (float)m_defaultFcParams.silenceThreshold;
Chris@151 232
Chris@156 233 desc.identifier = "metric";
Chris@156 234 desc.name = "Distance metric";
Chris@156 235 desc.description = "Metric for distance calculations.";
Chris@156 236 desc.minValue = 0;
Chris@157 237 desc.maxValue = 2;
Chris@156 238 desc.defaultValue = (int)m_defaultDParams.metric;
Chris@156 239 desc.isQuantized = true;
Chris@156 240 desc.quantizeStep = 1;
Chris@156 241 desc.valueNames.clear();
Chris@157 242 desc.valueNames.push_back("Manhattan");
Chris@156 243 desc.valueNames.push_back("Euclidean");
Chris@156 244 desc.valueNames.push_back("Cosine");
Chris@156 245 list.push_back(desc);
Chris@156 246 desc.valueNames.clear();
Chris@156 247
Chris@162 248 desc.identifier = "distnorm";
Chris@162 249 desc.name = "Distance normalisation";
Chris@162 250 desc.description = "Type of normalisation to use for distance metric";
Chris@162 251 desc.minValue = 0;
Chris@162 252 desc.maxValue = 2;
Chris@162 253 desc.defaultValue = (int)m_defaultDParams.norm;
Chris@162 254 desc.isQuantized = true;
Chris@162 255 desc.quantizeStep = 1;
Chris@162 256 desc.valueNames.clear();
Chris@162 257 desc.valueNames.push_back("None");
Chris@162 258 desc.valueNames.push_back("Sum of frames");
Chris@162 259 desc.valueNames.push_back("Log sum of frames");
Chris@162 260 list.push_back(desc);
Chris@162 261 desc.valueNames.clear();
Chris@162 262
Chris@162 263 desc.identifier = "silencethreshold";
Chris@162 264 desc.name = "Silence threshold";
Chris@162 265 desc.description = "Total frame energy threshold below which a feature will be regarded as silent";
Chris@162 266 desc.minValue = 0;
Chris@165 267 desc.maxValue = 0.1f;
Chris@165 268 desc.defaultValue = (float)m_defaultFcParams.silenceThreshold;
Chris@162 269 desc.isQuantized = false;
Chris@162 270 list.push_back(desc);
Chris@162 271
Chris@151 272 desc.identifier = "noise";
Chris@162 273 desc.name = "Add noise";
Chris@151 274 desc.description = "Whether to mix in a small constant white noise term when calculating feature distance. This can improve alignment against sources containing cleanly synthesised audio.";
Chris@151 275 desc.minValue = 0;
Chris@151 276 desc.maxValue = 1;
Chris@151 277 desc.defaultValue = (int)m_defaultDParams.noise;
Chris@151 278 desc.isQuantized = true;
Chris@151 279 desc.quantizeStep = 1;
Chris@151 280 list.push_back(desc);
Chris@136 281
Chris@25 282 desc.identifier = "gradientlimit";
Chris@162 283 desc.name = "Gradient limit";
Chris@18 284 desc.description = "Limit of number of frames that will be accepted from one source without a frame from the other source being accepted";
Chris@18 285 desc.minValue = 1;
Chris@18 286 desc.maxValue = 10;
Chris@18 287 desc.defaultValue = m_defaultParams.maxRunCount;
Chris@18 288 desc.isQuantized = true;
Chris@18 289 desc.quantizeStep = 1;
Chris@18 290 list.push_back(desc);
Chris@18 291
Chris@25 292 desc.identifier = "zonewidth";
Chris@162 293 desc.name = "Search zone width";
Chris@25 294 desc.description = "Width of the search zone (error margin) either side of the ongoing match position, in seconds";
Chris@25 295 desc.minValue = 1;
Chris@25 296 desc.maxValue = 60;
Chris@52 297 desc.defaultValue = (float)m_defaultParams.blockTime;
Chris@25 298 desc.isQuantized = true;
Chris@25 299 desc.quantizeStep = 1;
Chris@25 300 desc.unit = "s";
Chris@25 301 list.push_back(desc);
Chris@25 302
Chris@83 303 desc.identifier = "diagonalweight";
Chris@162 304 desc.name = "Diagonal weight";
Chris@83 305 desc.description = "Weight applied to cost of diagonal step relative to horizontal or vertical step. The default of 2.0 is good for gross tracking of quite different performances; closer to 1.0 produces a smoother path for performances more similar in tempo";
Chris@83 306 desc.minValue = 1.0;
Chris@86 307 desc.maxValue = 2.0;
Chris@138 308 desc.defaultValue = (float)m_defaultParams.diagonalWeight;
Chris@83 309 desc.isQuantized = false;
Chris@83 310 desc.unit = "";
Chris@83 311 list.push_back(desc);
Chris@83 312
Chris@32 313 desc.identifier = "smooth";
Chris@162 314 desc.name = "Use path smoothing";
Chris@138 315 desc.description = "Smooth the path by replacing steps with diagonals. (This was enabled by default in earlier versions of the MATCH plugin, but the default now is to produce an un-smoothed path.)";
Chris@32 316 desc.minValue = 0;
Chris@32 317 desc.maxValue = 1;
Chris@138 318 desc.defaultValue = 0;
Chris@32 319 desc.isQuantized = true;
Chris@32 320 desc.quantizeStep = 1;
Chris@32 321 desc.unit = "";
Chris@32 322 list.push_back(desc);
Chris@32 323
Chris@162 324 desc.identifier = "serialise";
Chris@162 325 desc.name = "Serialise plugin invocations";
Chris@162 326 desc.description = "Reduce potential memory load at the expense of multiprocessor performance by serialising multi-threaded plugin runs";
Chris@162 327 desc.minValue = 0;
Chris@162 328 desc.maxValue = 1;
Chris@162 329 desc.defaultValue = 0;
Chris@162 330 desc.isQuantized = true;
Chris@162 331 desc.quantizeStep = 1;
Chris@161 332 list.push_back(desc);
Chris@161 333
cannam@0 334 return list;
cannam@0 335 }
cannam@0 336
cannam@0 337 float
cannam@0 338 MatchVampPlugin::getParameter(std::string name) const
cannam@0 339 {
cannam@0 340 if (name == "serialise") {
cannam@0 341 return m_serialise ? 1.0 : 0.0;
Chris@18 342 } else if (name == "framenorm") {
Chris@103 343 return (int)m_fcParams.norm;
Chris@18 344 } else if (name == "distnorm") {
Chris@143 345 return (int)m_dParams.norm;
Chris@18 346 } else if (name == "usespecdiff") {
Chris@103 347 return (int)m_fcParams.order;
Chris@18 348 } else if (name == "usechroma") {
Chris@38 349 return m_feParams.useChromaFrequencyMap ? 1.0 : 0.0;
Chris@25 350 } else if (name == "gradientlimit") {
Chris@18 351 return m_params.maxRunCount;
Chris@83 352 } else if (name == "diagonalweight") {
Chris@83 353 return m_params.diagonalWeight;
Chris@25 354 } else if (name == "zonewidth") {
Chris@52 355 return (float)m_params.blockTime;
Chris@32 356 } else if (name == "smooth") {
Chris@32 357 return m_smooth ? 1.0 : 0.0;
Chris@136 358 } else if (name == "silencethreshold") {
Chris@148 359 return (float)m_fcParams.silenceThreshold;
Chris@156 360 } else if (name == "metric") {
Chris@156 361 return (int)m_dParams.metric;
Chris@151 362 } else if (name == "noise") {
Chris@151 363 return m_dParams.noise;
Chris@161 364 } else if (name == "freq1") {
Chris@161 365 return (float)m_feParams.referenceFrequency;
Chris@161 366 } else if (name == "freq2") {
Chris@161 367 return (float)m_secondReferenceFrequency;
Chris@177 368 } else if (name == "minfreq") {
Chris@177 369 return (float)m_feParams.minFrequency;
Chris@177 370 } else if (name == "maxfreq") {
Chris@177 371 return (float)m_feParams.maxFrequency;
cannam@0 372 }
Chris@18 373
cannam@0 374 return 0.0;
cannam@0 375 }
cannam@0 376
cannam@0 377 void
cannam@0 378 MatchVampPlugin::setParameter(std::string name, float value)
cannam@0 379 {
cannam@0 380 if (name == "serialise") {
cannam@0 381 m_serialise = (value > 0.5);
Chris@18 382 } else if (name == "framenorm") {
Chris@103 383 m_fcParams.norm = (FeatureConditioner::Normalisation)(int(value + 0.1));
Chris@18 384 } else if (name == "distnorm") {
Chris@143 385 m_dParams.norm = (DistanceMetric::DistanceNormalisation)(int(value + 0.1));
Chris@18 386 } else if (name == "usespecdiff") {
Chris@103 387 m_fcParams.order = (FeatureConditioner::OutputOrder)(int(value + 0.1));
Chris@18 388 } else if (name == "usechroma") {
Chris@38 389 m_feParams.useChromaFrequencyMap = (value > 0.5);
Chris@25 390 } else if (name == "gradientlimit") {
Chris@18 391 m_params.maxRunCount = int(value + 0.1);
Chris@83 392 } else if (name == "diagonalweight") {
Chris@83 393 m_params.diagonalWeight = value;
Chris@25 394 } else if (name == "zonewidth") {
Chris@25 395 m_params.blockTime = value;
Chris@32 396 } else if (name == "smooth") {
Chris@32 397 m_smooth = (value > 0.5);
Chris@136 398 } else if (name == "silencethreshold") {
Chris@136 399 m_fcParams.silenceThreshold = value;
Chris@156 400 } else if (name == "metric") {
Chris@156 401 m_dParams.metric = (DistanceMetric::Metric)(int(value + 0.1));
Chris@151 402 } else if (name == "noise") {
Chris@151 403 m_dParams.noise = (DistanceMetric::NoiseAddition)(int(value + 0.1));
Chris@161 404 } else if (name == "freq1") {
Chris@161 405 m_feParams.referenceFrequency = value;
Chris@161 406 } else if (name == "freq2") {
Chris@161 407 m_secondReferenceFrequency = value;
Chris@177 408 } else if (name == "minfreq") {
Chris@177 409 m_feParams.minFrequency = value;
Chris@177 410 } else if (name == "maxfreq") {
Chris@177 411 m_feParams.maxFrequency = value;
cannam@0 412 }
cannam@0 413 }
cannam@0 414
cannam@0 415 size_t
cannam@0 416 MatchVampPlugin::getPreferredStepSize() const
cannam@0 417 {
Chris@52 418 return int(m_inputSampleRate * defaultStepTime + 0.001);
cannam@0 419 }
cannam@0 420
cannam@0 421 size_t
cannam@0 422 MatchVampPlugin::getPreferredBlockSize() const
cannam@0 423 {
Chris@15 424 return 2048;
cannam@0 425 }
cannam@0 426
cannam@0 427 void
Chris@17 428 MatchVampPlugin::createMatchers()
cannam@0 429 {
Chris@17 430 m_params.hopTime = m_stepTime;
Chris@38 431 m_feParams.fftSize = m_blockSize;
Chris@107 432
Chris@161 433 m_pipeline = new MatchPipeline(m_feParams, m_fcParams, m_dParams, m_params,
Chris@161 434 m_secondReferenceFrequency);
cannam@0 435 }
cannam@0 436
cannam@0 437 bool
cannam@0 438 MatchVampPlugin::initialise(size_t channels, size_t stepSize, size_t blockSize)
cannam@0 439 {
Chris@10 440 if (m_inputSampleRate < sampleRateMin) {
Chris@10 441 std::cerr << "MatchVampPlugin::MatchVampPlugin: input sample rate "
Chris@10 442 << m_inputSampleRate << " < min supported rate "
Chris@10 443 << sampleRateMin << std::endl;
Chris@10 444 return false;
Chris@10 445 }
cannam@0 446 if (channels < getMinChannelCount() ||
cannam@0 447 channels > getMaxChannelCount()) return false;
cannam@1 448 if (stepSize > blockSize/2 ||
cannam@0 449 blockSize != getPreferredBlockSize()) return false;
Chris@15 450
cannam@6 451 m_stepSize = stepSize;
Chris@15 452 m_stepTime = float(stepSize) / m_inputSampleRate;
Chris@15 453 m_blockSize = blockSize;
Chris@15 454
Chris@15 455 createMatchers();
cannam@0 456 m_begin = true;
cannam@0 457 m_locked = false;
Chris@15 458
cannam@0 459 return true;
cannam@0 460 }
cannam@0 461
cannam@0 462 void
cannam@0 463 MatchVampPlugin::reset()
cannam@0 464 {
Chris@107 465 delete m_pipeline;
Chris@107 466 m_pipeline = 0;
Chris@74 467 m_frameNo = 0;
cannam@6 468 createMatchers();
cannam@6 469 m_begin = true;
cannam@6 470 m_locked = false;
cannam@0 471 }
cannam@0 472
cannam@0 473 MatchVampPlugin::OutputList
cannam@0 474 MatchVampPlugin::getOutputDescriptors() const
cannam@0 475 {
cannam@0 476 OutputList list;
cannam@0 477
Chris@52 478 float outRate = 1.0f / m_stepTime;
cannam@0 479
cannam@0 480 OutputDescriptor desc;
cannam@0 481 desc.identifier = "path";
cannam@0 482 desc.name = "Path";
cannam@0 483 desc.description = "Alignment path";
cannam@0 484 desc.unit = "";
cannam@0 485 desc.hasFixedBinCount = true;
cannam@0 486 desc.binCount = 1;
cannam@0 487 desc.hasKnownExtents = false;
cannam@0 488 desc.isQuantized = true;
cannam@0 489 desc.quantizeStep = 1;
cannam@0 490 desc.sampleType = OutputDescriptor::VariableSampleRate;
cannam@0 491 desc.sampleRate = outRate;
Chris@16 492 m_pathOutNo = list.size();
cannam@0 493 list.push_back(desc);
cannam@0 494
cannam@0 495 desc.identifier = "a_b";
cannam@0 496 desc.name = "A-B Timeline";
cannam@0 497 desc.description = "Timing in performance B corresponding to moments in performance A";
cannam@0 498 desc.unit = "sec";
cannam@0 499 desc.hasFixedBinCount = true;
cannam@0 500 desc.binCount = 1;
cannam@0 501 desc.hasKnownExtents = false;
cannam@0 502 desc.isQuantized = false;
cannam@0 503 desc.sampleType = OutputDescriptor::VariableSampleRate;
cannam@0 504 desc.sampleRate = outRate;
Chris@16 505 m_abOutNo = list.size();
cannam@0 506 list.push_back(desc);
cannam@0 507
cannam@0 508 desc.identifier = "b_a";
cannam@0 509 desc.name = "B-A Timeline";
cannam@0 510 desc.description = "Timing in performance A corresponding to moments in performance B";
cannam@0 511 desc.unit = "sec";
cannam@0 512 desc.hasFixedBinCount = true;
cannam@0 513 desc.binCount = 1;
cannam@0 514 desc.hasKnownExtents = false;
cannam@0 515 desc.isQuantized = false;
cannam@0 516 desc.sampleType = OutputDescriptor::VariableSampleRate;
cannam@0 517 desc.sampleRate = outRate;
Chris@16 518 m_baOutNo = list.size();
cannam@0 519 list.push_back(desc);
cannam@0 520
cannam@0 521 desc.identifier = "a_b_divergence";
cannam@0 522 desc.name = "A-B Divergence";
cannam@0 523 desc.description = "Difference between timings in performances A and B";
cannam@0 524 desc.unit = "sec";
cannam@0 525 desc.hasFixedBinCount = true;
cannam@0 526 desc.binCount = 1;
cannam@0 527 desc.hasKnownExtents = false;
cannam@0 528 desc.isQuantized = false;
cannam@0 529 desc.sampleType = OutputDescriptor::VariableSampleRate;
cannam@0 530 desc.sampleRate = outRate;
Chris@16 531 m_abDivOutNo = list.size();
cannam@0 532 list.push_back(desc);
cannam@0 533
cannam@0 534 desc.identifier = "a_b_temporatio";
cannam@0 535 desc.name = "A-B Tempo Ratio";
cannam@0 536 desc.description = "Ratio of tempi between performances A and B";
cannam@0 537 desc.unit = "";
cannam@0 538 desc.hasFixedBinCount = true;
cannam@0 539 desc.binCount = 1;
cannam@0 540 desc.hasKnownExtents = false;
cannam@0 541 desc.isQuantized = false;
cannam@0 542 desc.sampleType = OutputDescriptor::VariableSampleRate;
cannam@0 543 desc.sampleRate = outRate;
Chris@16 544 m_abRatioOutNo = list.size();
cannam@0 545 list.push_back(desc);
cannam@0 546
Chris@38 547 int featureSize = FeatureExtractor(m_feParams).getFeatureSize();
Chris@38 548
Chris@15 549 desc.identifier = "a_features";
Chris@140 550 desc.name = "Raw A Features";
Chris@15 551 desc.description = "Spectral features extracted from performance A";
Chris@15 552 desc.unit = "";
Chris@15 553 desc.hasFixedBinCount = true;
Chris@38 554 desc.binCount = featureSize;
Chris@15 555 desc.hasKnownExtents = false;
Chris@15 556 desc.isQuantized = false;
Chris@16 557 desc.sampleType = OutputDescriptor::FixedSampleRate;
Chris@15 558 desc.sampleRate = outRate;
Chris@16 559 m_aFeaturesOutNo = list.size();
Chris@16 560 list.push_back(desc);
Chris@16 561
Chris@16 562 desc.identifier = "b_features";
Chris@140 563 desc.name = "Raw B Features";
Chris@16 564 desc.description = "Spectral features extracted from performance B";
Chris@16 565 desc.unit = "";
Chris@16 566 desc.hasFixedBinCount = true;
Chris@38 567 desc.binCount = featureSize;
Chris@16 568 desc.hasKnownExtents = false;
Chris@16 569 desc.isQuantized = false;
Chris@16 570 desc.sampleType = OutputDescriptor::FixedSampleRate;
Chris@16 571 desc.sampleRate = outRate;
Chris@16 572 m_bFeaturesOutNo = list.size();
Chris@15 573 list.push_back(desc);
Chris@15 574
Chris@140 575 desc.identifier = "a_cfeatures";
Chris@140 576 desc.name = "Conditioned A Features";
Chris@140 577 desc.description = "Spectral features extracted from performance A, after normalisation and conditioning";
Chris@140 578 desc.unit = "";
Chris@140 579 desc.hasFixedBinCount = true;
Chris@140 580 desc.binCount = featureSize;
Chris@140 581 desc.hasKnownExtents = false;
Chris@140 582 desc.isQuantized = false;
Chris@140 583 desc.sampleType = OutputDescriptor::FixedSampleRate;
Chris@140 584 desc.sampleRate = outRate;
Chris@140 585 m_caFeaturesOutNo = list.size();
Chris@140 586 list.push_back(desc);
Chris@140 587
Chris@140 588 desc.identifier = "b_cfeatures";
Chris@140 589 desc.name = "Conditioned B Features";
Chris@140 590 desc.description = "Spectral features extracted from performance B, after norrmalisation and conditioning";
Chris@140 591 desc.unit = "";
Chris@140 592 desc.hasFixedBinCount = true;
Chris@140 593 desc.binCount = featureSize;
Chris@140 594 desc.hasKnownExtents = false;
Chris@140 595 desc.isQuantized = false;
Chris@140 596 desc.sampleType = OutputDescriptor::FixedSampleRate;
Chris@140 597 desc.sampleRate = outRate;
Chris@140 598 m_cbFeaturesOutNo = list.size();
Chris@140 599 list.push_back(desc);
Chris@140 600
Chris@163 601 desc.identifier = "overall_cost";
Chris@163 602 desc.name = "Overall Cost";
Chris@163 603 desc.description = "Normalised overall path cost for the cheapest path";
Chris@163 604 desc.unit = "";
Chris@163 605 desc.hasFixedBinCount = true;
Chris@163 606 desc.binCount = 1;
Chris@163 607 desc.hasKnownExtents = false;
Chris@163 608 desc.isQuantized = false;
Chris@163 609 desc.sampleType = OutputDescriptor::FixedSampleRate;
Chris@163 610 desc.sampleRate = 1;
Chris@163 611 m_overallCostOutNo = list.size();
Chris@163 612 list.push_back(desc);
Chris@163 613
cannam@0 614 return list;
cannam@0 615 }
cannam@0 616
cannam@0 617 MatchVampPlugin::FeatureSet
cannam@0 618 MatchVampPlugin::process(const float *const *inputBuffers,
cannam@0 619 Vamp::RealTime timestamp)
cannam@0 620 {
cannam@0 621 if (m_begin) {
cannam@0 622 if (!m_locked && m_serialise) {
cannam@0 623 m_locked = true;
cannam@0 624 #ifdef _WIN32
cannam@0 625 WaitForSingleObject(m_serialisingMutex, INFINITE);
cannam@0 626 #else
cannam@0 627 pthread_mutex_lock(&m_serialisingMutex);
cannam@0 628 #endif
cannam@0 629 }
Chris@10 630 m_startTime = timestamp;
cannam@0 631 m_begin = false;
cannam@0 632 }
cannam@0 633
cannam@0 634 // std::cerr << timestamp.toString();
cannam@0 635
Chris@107 636 m_pipeline->feedFrequencyDomainAudio(inputBuffers[0], inputBuffers[1]);
Chris@74 637
Chris@140 638 FeatureSet returnFeatures;
Chris@140 639
Chris@107 640 vector<double> f1, f2;
Chris@140 641 m_pipeline->extractFeatures(f1, f2);
Chris@16 642
Chris@140 643 vector<double> cf1, cf2;
Chris@140 644 m_pipeline->extractConditionedFeatures(cf1, cf2);
Chris@16 645
Chris@16 646 Feature f;
Chris@16 647 f.hasTimestamp = false;
Chris@16 648
Chris@74 649 f.values.clear();
Chris@74 650 for (int j = 0; j < (int)f1.size(); ++j) {
Chris@74 651 f.values.push_back(float(f1[j]));
Chris@16 652 }
Chris@74 653 returnFeatures[m_aFeaturesOutNo].push_back(f);
Chris@16 654
Chris@74 655 f.values.clear();
Chris@74 656 for (int j = 0; j < (int)f2.size(); ++j) {
Chris@74 657 f.values.push_back(float(f2[j]));
Chris@16 658 }
Chris@74 659 returnFeatures[m_bFeaturesOutNo].push_back(f);
cannam@0 660
Chris@140 661 f.values.clear();
Chris@140 662 for (int j = 0; j < (int)cf1.size(); ++j) {
Chris@140 663 f.values.push_back(float(cf1[j]));
Chris@140 664 }
Chris@140 665 returnFeatures[m_caFeaturesOutNo].push_back(f);
Chris@140 666
Chris@140 667 f.values.clear();
Chris@140 668 for (int j = 0; j < (int)cf2.size(); ++j) {
Chris@140 669 f.values.push_back(float(cf2[j]));
Chris@140 670 }
Chris@140 671 returnFeatures[m_cbFeaturesOutNo].push_back(f);
Chris@140 672
cannam@0 673 // std::cerr << ".";
cannam@0 674 // std::cerr << std::endl;
cannam@0 675
Chris@74 676 ++m_frameNo;
Chris@74 677
Chris@16 678 return returnFeatures;
cannam@0 679 }
cannam@0 680
cannam@0 681 MatchVampPlugin::FeatureSet
cannam@0 682 MatchVampPlugin::getRemainingFeatures()
cannam@0 683 {
Chris@107 684 m_pipeline->finish();
Chris@74 685
Chris@63 686 FeatureSet returnFeatures;
Chris@63 687
cannam@0 688 std::vector<int> pathx;
cannam@0 689 std::vector<int> pathy;
Chris@155 690 int len = m_pipeline->retrievePath(m_smooth, pathx, pathy);
cannam@0 691
Chris@173 692 double cost = m_pipeline->getOverallCost();
Chris@163 693 Feature costFeature;
Chris@163 694 costFeature.hasTimestamp = false;
Chris@163 695 costFeature.values.push_back((float)cost);
Chris@163 696 returnFeatures[m_overallCostOutNo].push_back(costFeature);
Chris@163 697
cannam@0 698 int prevx = 0;
cannam@0 699 int prevy = 0;
cannam@0 700
Chris@30 701 for (int i = 0; i < len; ++i) {
cannam@0 702
cannam@0 703 int x = pathx[i];
cannam@0 704 int y = pathy[i];
cannam@0 705
cannam@0 706 Vamp::RealTime xt = Vamp::RealTime::frame2RealTime
Chris@15 707 (x * m_stepSize, lrintf(m_inputSampleRate));
cannam@0 708 Vamp::RealTime yt = Vamp::RealTime::frame2RealTime
Chris@15 709 (y * m_stepSize, lrintf(m_inputSampleRate));
cannam@0 710
cannam@0 711 Feature feature;
cannam@0 712 feature.hasTimestamp = true;
Chris@10 713 feature.timestamp = m_startTime + xt;
cannam@0 714 feature.values.clear();
Chris@52 715 feature.values.push_back(float(yt.sec + double(yt.nsec)/1.0e9));
Chris@16 716 returnFeatures[m_pathOutNo].push_back(feature);
cannam@0 717
cannam@0 718 if (x != prevx) {
cannam@0 719
cannam@0 720 feature.hasTimestamp = true;
Chris@10 721 feature.timestamp = m_startTime + xt;
cannam@0 722 feature.values.clear();
Chris@52 723 feature.values.push_back(float(yt.sec + yt.msec()/1000.0));
Chris@16 724 returnFeatures[m_abOutNo].push_back(feature);
cannam@0 725
cannam@0 726 Vamp::RealTime diff = yt - xt;
cannam@0 727 feature.values.clear();
Chris@52 728 feature.values.push_back(float(diff.sec + diff.msec()/1000.0));
Chris@16 729 returnFeatures[m_abDivOutNo].push_back(feature);
cannam@0 730
cannam@0 731 if (i > 0) {
cannam@0 732 int lookback = 100; //!!! arbitrary
cannam@0 733 if (lookback > i) lookback = i;
cannam@0 734 int xdiff = x - pathx[i-lookback];
cannam@0 735 int ydiff = y - pathy[i-lookback];
cannam@0 736 if (xdiff != 0 && ydiff != 0) {
cannam@0 737 float ratio = float(ydiff)/float(xdiff);
cannam@0 738 if (ratio < 8 && ratio > (1.0/8)) { //!!! just for now, since we aren't dealing properly with silence yet
cannam@0 739 feature.values.clear();
cannam@0 740 feature.values.push_back(ratio);
Chris@16 741 returnFeatures[m_abRatioOutNo].push_back(feature);
cannam@0 742 }
cannam@0 743 }
cannam@0 744 }
cannam@0 745 }
cannam@0 746
cannam@0 747 if (y != prevy) {
cannam@0 748 feature.hasTimestamp = true;
Chris@10 749 feature.timestamp = m_startTime + yt;
cannam@0 750 feature.values.clear();
Chris@52 751 feature.values.push_back(float(xt.sec + xt.msec()/1000.0));
Chris@16 752 returnFeatures[m_baOutNo].push_back(feature);
cannam@0 753 }
cannam@0 754
cannam@0 755 prevx = x;
cannam@0 756 prevy = y;
cannam@0 757 }
cannam@0 758
Chris@107 759 delete m_pipeline;
Chris@107 760 m_pipeline = 0;
cannam@0 761
cannam@0 762 if (m_locked) {
cannam@0 763 #ifdef _WIN32
cannam@0 764 ReleaseMutex(m_serialisingMutex);
cannam@0 765 #else
cannam@0 766 pthread_mutex_unlock(&m_serialisingMutex);
cannam@0 767 #endif
cannam@0 768 m_locked = false;
cannam@0 769 }
cannam@0 770
cannam@0 771 return returnFeatures;
cannam@0 772
cannam@0 773
cannam@0 774 /*
Chris@30 775 for (int i = 0; i < len; ++i) {
cannam@0 776 std::cerr << i << ": [" << pathx[i] << "," << pathy[i] << "]" << std::endl;
cannam@0 777 }
cannam@0 778
cannam@0 779 std::cerr << std::endl;
cannam@0 780 std::cerr << "File: A" << std::endl;
cannam@0 781 std::cerr << "Marks: -1" << std::endl;
cannam@0 782 std::cerr << "FixedPoints: true 0" << std::endl;
cannam@0 783 std::cerr << "0" << std::endl;
cannam@0 784 std::cerr << "0" << std::endl;
cannam@0 785 std::cerr << "0" << std::endl;
cannam@0 786 std::cerr << "0" << std::endl;
cannam@0 787 std::cerr << "File: B" << std::endl;
cannam@0 788 std::cerr << "Marks: 0" << std::endl;
cannam@0 789 std::cerr << "FixedPoints: true 0" << std::endl;
cannam@0 790 std::cerr << "0.02" << std::endl;
cannam@0 791 std::cerr << "0.02" << std::endl;
cannam@0 792
Chris@30 793 std::cerr << len << std::endl;
Chris@30 794 for (int i = 0; i < len; ++i) {
cannam@0 795 std::cerr << pathx[i] << std::endl;
cannam@0 796 }
cannam@0 797
Chris@30 798 std::cerr << len << std::endl;
Chris@30 799 for (int i = 0; i < len; ++i) {
cannam@0 800 std::cerr << pathy[i] << std::endl;
cannam@0 801 }
cannam@0 802 */
cannam@0 803 }
cannam@0 804
cannam@0 805 static Vamp::PluginAdapter<MatchVampPlugin> mvpAdapter;
cannam@0 806
cannam@0 807 const VampPluginDescriptor *vampGetPluginDescriptor(unsigned int version,
cannam@0 808 unsigned int index)
cannam@0 809 {
cannam@0 810 if (version < 1) return 0;
cannam@0 811
cannam@0 812 switch (index) {
cannam@0 813 case 0: return mvpAdapter.getDescriptor();
cannam@0 814 default: return 0;
cannam@0 815 }
cannam@0 816 }