Chris@0
|
1
|
Chris@0
|
2 #include "TuningDifference.h"
|
Chris@0
|
3
|
Chris@1
|
4 #include <iostream>
|
Chris@1
|
5
|
Chris@4
|
6 #include <cmath>
|
Chris@4
|
7 #include <cstdio>
|
Chris@4
|
8
|
Chris@13
|
9 #include <algorithm>
|
Chris@1
|
10
|
Chris@13
|
11 using namespace std;
|
Chris@13
|
12
|
Chris@13
|
13 static double pitchToFrequency(int pitch,
|
Chris@13
|
14 double centsOffset = 0.,
|
Chris@13
|
15 double concertA = 440.)
|
Chris@13
|
16 {
|
Chris@13
|
17 double p = double(pitch) + (centsOffset / 100.);
|
Chris@13
|
18 return concertA * pow(2.0, (p - 69.0) / 12.0);
|
Chris@13
|
19 }
|
Chris@13
|
20
|
Chris@13
|
21 static double frequencyForCentsAbove440(double cents)
|
Chris@13
|
22 {
|
Chris@13
|
23 return pitchToFrequency(69, cents, 440.);
|
Chris@13
|
24 }
|
Chris@5
|
25
|
Chris@0
|
26 TuningDifference::TuningDifference(float inputSampleRate) :
|
Chris@13
|
27 Plugin(inputSampleRate),
|
Chris@13
|
28 m_bpo(60),
|
Chris@13
|
29 m_refCQ(new CQSpectrogram(paramsForTuningFrequency(440.),
|
Chris@13
|
30 CQSpectrogram::InterpolateHold))
|
Chris@0
|
31 {
|
Chris@0
|
32 }
|
Chris@0
|
33
|
Chris@0
|
34 TuningDifference::~TuningDifference()
|
Chris@0
|
35 {
|
Chris@0
|
36 }
|
Chris@0
|
37
|
Chris@0
|
38 string
|
Chris@0
|
39 TuningDifference::getIdentifier() const
|
Chris@0
|
40 {
|
Chris@1
|
41 return "tuning-difference";
|
Chris@0
|
42 }
|
Chris@0
|
43
|
Chris@0
|
44 string
|
Chris@0
|
45 TuningDifference::getName() const
|
Chris@0
|
46 {
|
Chris@1
|
47 return "Tuning Difference";
|
Chris@0
|
48 }
|
Chris@0
|
49
|
Chris@0
|
50 string
|
Chris@0
|
51 TuningDifference::getDescription() const
|
Chris@0
|
52 {
|
Chris@0
|
53 // Return something helpful here!
|
Chris@0
|
54 return "";
|
Chris@0
|
55 }
|
Chris@0
|
56
|
Chris@0
|
57 string
|
Chris@0
|
58 TuningDifference::getMaker() const
|
Chris@0
|
59 {
|
Chris@0
|
60 // Your name here
|
Chris@0
|
61 return "";
|
Chris@0
|
62 }
|
Chris@0
|
63
|
Chris@0
|
64 int
|
Chris@0
|
65 TuningDifference::getPluginVersion() const
|
Chris@0
|
66 {
|
Chris@0
|
67 // Increment this each time you release a version that behaves
|
Chris@0
|
68 // differently from the previous one
|
Chris@0
|
69 return 1;
|
Chris@0
|
70 }
|
Chris@0
|
71
|
Chris@0
|
72 string
|
Chris@0
|
73 TuningDifference::getCopyright() const
|
Chris@0
|
74 {
|
Chris@0
|
75 // This function is not ideally named. It does not necessarily
|
Chris@0
|
76 // need to say who made the plugin -- getMaker does that -- but it
|
Chris@0
|
77 // should indicate the terms under which it is distributed. For
|
Chris@0
|
78 // example, "Copyright (year). All Rights Reserved", or "GPL"
|
Chris@0
|
79 return "";
|
Chris@0
|
80 }
|
Chris@0
|
81
|
Chris@0
|
82 TuningDifference::InputDomain
|
Chris@0
|
83 TuningDifference::getInputDomain() const
|
Chris@0
|
84 {
|
Chris@13
|
85 return TimeDomain;
|
Chris@0
|
86 }
|
Chris@0
|
87
|
Chris@0
|
88 size_t
|
Chris@0
|
89 TuningDifference::getPreferredBlockSize() const
|
Chris@0
|
90 {
|
Chris@13
|
91 return 0;
|
Chris@0
|
92 }
|
Chris@0
|
93
|
Chris@0
|
94 size_t
|
Chris@0
|
95 TuningDifference::getPreferredStepSize() const
|
Chris@0
|
96 {
|
Chris@1
|
97 return 0;
|
Chris@0
|
98 }
|
Chris@0
|
99
|
Chris@0
|
100 size_t
|
Chris@0
|
101 TuningDifference::getMinChannelCount() const
|
Chris@0
|
102 {
|
Chris@1
|
103 return 2;
|
Chris@0
|
104 }
|
Chris@0
|
105
|
Chris@0
|
106 size_t
|
Chris@0
|
107 TuningDifference::getMaxChannelCount() const
|
Chris@0
|
108 {
|
Chris@1
|
109 return 2;
|
Chris@0
|
110 }
|
Chris@0
|
111
|
Chris@0
|
112 TuningDifference::ParameterList
|
Chris@0
|
113 TuningDifference::getParameterDescriptors() const
|
Chris@0
|
114 {
|
Chris@0
|
115 ParameterList list;
|
Chris@0
|
116 return list;
|
Chris@0
|
117 }
|
Chris@0
|
118
|
Chris@0
|
119 float
|
Chris@1
|
120 TuningDifference::getParameter(string) const
|
Chris@0
|
121 {
|
Chris@0
|
122 return 0;
|
Chris@0
|
123 }
|
Chris@0
|
124
|
Chris@0
|
125 void
|
Chris@1
|
126 TuningDifference::setParameter(string, float)
|
Chris@0
|
127 {
|
Chris@0
|
128 }
|
Chris@0
|
129
|
Chris@0
|
130 TuningDifference::ProgramList
|
Chris@0
|
131 TuningDifference::getPrograms() const
|
Chris@0
|
132 {
|
Chris@0
|
133 ProgramList list;
|
Chris@0
|
134 return list;
|
Chris@0
|
135 }
|
Chris@0
|
136
|
Chris@0
|
137 string
|
Chris@0
|
138 TuningDifference::getCurrentProgram() const
|
Chris@0
|
139 {
|
Chris@0
|
140 return ""; // no programs
|
Chris@0
|
141 }
|
Chris@0
|
142
|
Chris@0
|
143 void
|
Chris@1
|
144 TuningDifference::selectProgram(string)
|
Chris@0
|
145 {
|
Chris@0
|
146 }
|
Chris@0
|
147
|
Chris@0
|
148 TuningDifference::OutputList
|
Chris@0
|
149 TuningDifference::getOutputDescriptors() const
|
Chris@0
|
150 {
|
Chris@0
|
151 OutputList list;
|
Chris@0
|
152
|
Chris@1
|
153 OutputDescriptor d;
|
Chris@1
|
154 d.identifier = "cents";
|
Chris@1
|
155 d.name = "Tuning Difference";
|
Chris@1
|
156 d.description = "Difference in averaged frequency profile between channels 1 and 2, in cents. A positive value means channel 2 is higher.";
|
Chris@1
|
157 d.unit = "cents";
|
Chris@1
|
158 d.hasFixedBinCount = true;
|
Chris@1
|
159 d.binCount = 1;
|
Chris@1
|
160 d.hasKnownExtents = false;
|
Chris@1
|
161 d.isQuantized = false;
|
Chris@1
|
162 d.sampleType = OutputDescriptor::VariableSampleRate;
|
Chris@1
|
163 d.hasDuration = false;
|
Chris@13
|
164 m_outputs[d.identifier] = list.size();
|
Chris@1
|
165 list.push_back(d);
|
Chris@0
|
166
|
Chris@1
|
167 d.identifier = "tuningfreq";
|
Chris@1
|
168 d.name = "Relative Tuning Frequency";
|
Chris@1
|
169 d.description = "Tuning frequency of channel 2, if channel 1 is assumed to contain the same music as it at a tuning frequency of A=440Hz.";
|
Chris@4
|
170 d.unit = "hz";
|
Chris@1
|
171 d.hasFixedBinCount = true;
|
Chris@1
|
172 d.binCount = 1;
|
Chris@1
|
173 d.hasKnownExtents = false;
|
Chris@1
|
174 d.isQuantized = false;
|
Chris@1
|
175 d.sampleType = OutputDescriptor::VariableSampleRate;
|
Chris@1
|
176 d.hasDuration = false;
|
Chris@13
|
177 m_outputs[d.identifier] = list.size();
|
Chris@1
|
178 list.push_back(d);
|
Chris@1
|
179
|
Chris@13
|
180 d.identifier = "reffeature";
|
Chris@13
|
181 d.name = "Reference Feature";
|
Chris@13
|
182 d.description = "Chroma feature from reference audio.";
|
Chris@0
|
183 d.unit = "";
|
Chris@0
|
184 d.hasFixedBinCount = true;
|
Chris@13
|
185 d.binCount = m_bpo;
|
Chris@4
|
186 d.hasKnownExtents = false;
|
Chris@4
|
187 d.isQuantized = false;
|
Chris@4
|
188 d.sampleType = OutputDescriptor::FixedSampleRate;
|
Chris@5
|
189 d.sampleRate = 1;
|
Chris@4
|
190 d.hasDuration = false;
|
Chris@13
|
191 m_outputs[d.identifier] = list.size();
|
Chris@13
|
192 list.push_back(d);
|
Chris@13
|
193
|
Chris@13
|
194 d.identifier = "otherfeature";
|
Chris@13
|
195 d.name = "Other Feature";
|
Chris@13
|
196 d.description = "Chroma feature from other audio, before rotation.";
|
Chris@13
|
197 d.unit = "";
|
Chris@13
|
198 d.hasFixedBinCount = true;
|
Chris@13
|
199 d.binCount = m_bpo;
|
Chris@13
|
200 d.hasKnownExtents = false;
|
Chris@13
|
201 d.isQuantized = false;
|
Chris@13
|
202 d.sampleType = OutputDescriptor::FixedSampleRate;
|
Chris@13
|
203 d.sampleRate = 1;
|
Chris@13
|
204 d.hasDuration = false;
|
Chris@13
|
205 m_outputs[d.identifier] = list.size();
|
Chris@13
|
206 list.push_back(d);
|
Chris@13
|
207
|
Chris@13
|
208 d.identifier = "rotfeature";
|
Chris@13
|
209 d.name = "Other Feature at Rotated Frequency";
|
Chris@13
|
210 d.description = "Chroma feature from reference audio calculated with the tuning frequency obtained from rotation matching.";
|
Chris@13
|
211 d.unit = "";
|
Chris@13
|
212 d.hasFixedBinCount = true;
|
Chris@13
|
213 d.binCount = m_bpo;
|
Chris@13
|
214 d.hasKnownExtents = false;
|
Chris@13
|
215 d.isQuantized = false;
|
Chris@13
|
216 d.sampleType = OutputDescriptor::FixedSampleRate;
|
Chris@13
|
217 d.sampleRate = 1;
|
Chris@13
|
218 d.hasDuration = false;
|
Chris@13
|
219 m_outputs[d.identifier] = list.size();
|
Chris@4
|
220 list.push_back(d);
|
Chris@4
|
221
|
Chris@0
|
222 return list;
|
Chris@0
|
223 }
|
Chris@0
|
224
|
Chris@0
|
225 bool
|
Chris@0
|
226 TuningDifference::initialise(size_t channels, size_t stepSize, size_t blockSize)
|
Chris@0
|
227 {
|
Chris@0
|
228 if (channels < getMinChannelCount() ||
|
Chris@0
|
229 channels > getMaxChannelCount()) return false;
|
Chris@0
|
230
|
Chris@13
|
231 if (stepSize != blockSize) return false;
|
Chris@0
|
232
|
Chris@1
|
233 m_blockSize = blockSize;
|
Chris@1
|
234
|
Chris@1
|
235 reset();
|
Chris@1
|
236
|
Chris@0
|
237 return true;
|
Chris@0
|
238 }
|
Chris@0
|
239
|
Chris@0
|
240 void
|
Chris@0
|
241 TuningDifference::reset()
|
Chris@0
|
242 {
|
Chris@13
|
243 if (m_frameCount > 0) {
|
Chris@13
|
244 m_refCQ.reset(new CQSpectrogram(paramsForTuningFrequency(440.),
|
Chris@13
|
245 CQSpectrogram::InterpolateHold));
|
Chris@13
|
246 m_frameCount = 0;
|
Chris@13
|
247 }
|
Chris@13
|
248 m_refTotals = Chroma(m_refCQ->getTotalBins(), 0.0);
|
Chris@13
|
249 m_other.clear();
|
Chris@13
|
250 }
|
Chris@13
|
251
|
Chris@13
|
252 template<typename T>
|
Chris@13
|
253 void addTo(vector<T> &a, const vector<T> &b)
|
Chris@13
|
254 {
|
Chris@13
|
255 transform(a.begin(), a.end(), b.begin(), a.begin(), plus<T>());
|
Chris@13
|
256 }
|
Chris@13
|
257
|
Chris@13
|
258 template<typename T>
|
Chris@13
|
259 T distance(const vector<T> &a, const vector<T> &b)
|
Chris@13
|
260 {
|
Chris@13
|
261 return inner_product(a.begin(), a.end(), b.begin(), T(),
|
Chris@13
|
262 plus<T>(), [](T x, T y) { return fabs(x - y); });
|
Chris@13
|
263 }
|
Chris@13
|
264
|
Chris@13
|
265 TuningDifference::TFeature
|
Chris@13
|
266 TuningDifference::computeFeatureFromTotals(const Chroma &totals) const
|
Chris@13
|
267 {
|
Chris@13
|
268 TFeature feature(m_bpo);
|
Chris@13
|
269 double sum = 0.0;
|
Chris@13
|
270
|
Chris@13
|
271 for (int i = 0; i < (int)totals.size(); ++i) {
|
Chris@13
|
272 double value = totals[i] / m_frameCount;
|
Chris@13
|
273 feature[i % m_bpo] += value;
|
Chris@13
|
274 sum += value;
|
Chris@13
|
275 }
|
Chris@13
|
276
|
Chris@13
|
277 for (int i = 0; i < m_bpo; ++i) {
|
Chris@13
|
278 feature[i] /= sum;
|
Chris@13
|
279 }
|
Chris@13
|
280
|
Chris@13
|
281 cerr << "computeFeatureFromTotals: feature values:" << endl;
|
Chris@13
|
282 for (auto v: feature) cerr << v << " ";
|
Chris@13
|
283 cerr << endl;
|
Chris@13
|
284
|
Chris@13
|
285 return feature;
|
Chris@13
|
286 }
|
Chris@13
|
287
|
Chris@13
|
288 CQParameters
|
Chris@13
|
289 TuningDifference::paramsForTuningFrequency(double hz) const
|
Chris@13
|
290 {
|
Chris@13
|
291 return CQParameters(m_inputSampleRate,
|
Chris@13
|
292 pitchToFrequency(36, hz),
|
Chris@13
|
293 pitchToFrequency(96, hz),
|
Chris@13
|
294 m_bpo);
|
Chris@13
|
295 }
|
Chris@13
|
296
|
Chris@13
|
297 TuningDifference::TFeature
|
Chris@13
|
298 TuningDifference::computeFeatureFromSignal(const Signal &signal, double hz) const
|
Chris@13
|
299 {
|
Chris@13
|
300 CQSpectrogram cq(paramsForTuningFrequency(hz),
|
Chris@13
|
301 CQSpectrogram::InterpolateHold);
|
Chris@13
|
302
|
Chris@13
|
303 Chroma totals(m_refCQ->getTotalBins(), 0.0);
|
Chris@13
|
304
|
Chris@13
|
305 for (int i = 0; i < m_frameCount; ++i) {
|
Chris@13
|
306 Signal::const_iterator first = signal.begin() + i * m_blockSize;
|
Chris@13
|
307 Signal::const_iterator last = first + m_blockSize;
|
Chris@13
|
308 if (last > signal.end()) last = signal.end();
|
Chris@13
|
309 CQSpectrogram::RealSequence input(first, last);
|
Chris@13
|
310 input.resize(m_blockSize);
|
Chris@13
|
311 CQSpectrogram::RealBlock block = cq.process(input);
|
Chris@13
|
312 for (const auto &v: block) addTo(totals, v);
|
Chris@13
|
313 }
|
Chris@13
|
314
|
Chris@13
|
315 return computeFeatureFromTotals(totals);
|
Chris@0
|
316 }
|
Chris@0
|
317
|
Chris@0
|
318 TuningDifference::FeatureSet
|
Chris@13
|
319 TuningDifference::process(const float *const *inputBuffers, Vamp::RealTime)
|
Chris@0
|
320 {
|
Chris@13
|
321 CQSpectrogram::RealBlock block;
|
Chris@13
|
322 CQSpectrogram::RealSequence input;
|
Chris@13
|
323
|
Chris@13
|
324 input = CQSpectrogram::RealSequence
|
Chris@13
|
325 (inputBuffers[0], inputBuffers[0] + m_blockSize);
|
Chris@13
|
326 block = m_refCQ->process(input);
|
Chris@13
|
327 for (const auto &v: block) addTo(m_refTotals, v);
|
Chris@13
|
328
|
Chris@13
|
329 m_other.insert(m_other.end(),
|
Chris@13
|
330 inputBuffers[1], inputBuffers[1] + m_blockSize);
|
Chris@1
|
331
|
Chris@1
|
332 ++m_frameCount;
|
Chris@0
|
333 return FeatureSet();
|
Chris@0
|
334 }
|
Chris@0
|
335
|
Chris@13
|
336 double
|
Chris@13
|
337 TuningDifference::featureDistance(const TFeature &other, int rotation) const
|
Chris@13
|
338 {
|
Chris@13
|
339 if (rotation == 0) {
|
Chris@13
|
340 return distance(m_refFeature, other);
|
Chris@13
|
341 } else {
|
Chris@13
|
342 TFeature r(other);
|
Chris@13
|
343 if (rotation > 0) {
|
Chris@13
|
344 rotate(r.begin(), r.begin() + rotation, r.end());
|
Chris@13
|
345 } else {
|
Chris@13
|
346 rotate(r.begin(), r.end() + rotation, r.end());
|
Chris@13
|
347 }
|
Chris@13
|
348 return distance(m_refFeature, r);
|
Chris@13
|
349 }
|
Chris@13
|
350 }
|
Chris@13
|
351
|
Chris@13
|
352 int
|
Chris@13
|
353 TuningDifference::findBestRotation(const TFeature &other) const
|
Chris@13
|
354 {
|
Chris@13
|
355 map<double, int> dists;
|
Chris@13
|
356
|
Chris@13
|
357 int maxSemis = 6;
|
Chris@13
|
358 int maxRotation = (m_bpo * maxSemis) / 12;
|
Chris@13
|
359
|
Chris@13
|
360 for (int r = -maxRotation; r <= maxRotation; ++r) {
|
Chris@13
|
361 double dist = featureDistance(other, r);
|
Chris@13
|
362 dists[dist] = r;
|
Chris@13
|
363 cerr << "rotation " << r << ": score " << dist << endl;
|
Chris@13
|
364 }
|
Chris@13
|
365
|
Chris@13
|
366 int best = dists.begin()->second;
|
Chris@13
|
367
|
Chris@13
|
368 cerr << "best is " << best << endl;
|
Chris@13
|
369 return best;
|
Chris@13
|
370 }
|
Chris@13
|
371
|
Chris@0
|
372 TuningDifference::FeatureSet
|
Chris@0
|
373 TuningDifference::getRemainingFeatures()
|
Chris@0
|
374 {
|
Chris@13
|
375 FeatureSet fs;
|
Chris@13
|
376 if (m_frameCount == 0) return fs;
|
Chris@13
|
377
|
Chris@13
|
378 m_refFeature = computeFeatureFromTotals(m_refTotals);
|
Chris@13
|
379 TFeature otherFeature = computeFeatureFromSignal(m_other, 440.);
|
Chris@1
|
380
|
Chris@1
|
381 Feature f;
|
Chris@1
|
382
|
Chris@4
|
383 f.values.clear();
|
Chris@13
|
384 for (auto v: m_refFeature) f.values.push_back(v);
|
Chris@13
|
385 fs[m_outputs["reffeature"]].push_back(f);
|
Chris@4
|
386
|
Chris@4
|
387 f.values.clear();
|
Chris@13
|
388 for (auto v: otherFeature) f.values.push_back(v);
|
Chris@13
|
389 fs[m_outputs["otherfeature"]].push_back(f);
|
Chris@13
|
390
|
Chris@13
|
391 int rotation = findBestRotation(otherFeature);
|
Chris@13
|
392
|
Chris@13
|
393 int coarseCents = -(rotation * 100) / (m_bpo / 12);
|
Chris@13
|
394
|
Chris@13
|
395 cerr << "rotation " << rotation << " -> cents " << coarseCents << endl;
|
Chris@13
|
396
|
Chris@13
|
397 double coarseHz = frequencyForCentsAbove440(coarseCents);
|
Chris@13
|
398
|
Chris@13
|
399 TFeature coarseFeature = computeFeatureFromSignal(m_other, coarseHz);
|
Chris@13
|
400 double coarseScore = featureDistance(coarseFeature);
|
Chris@13
|
401
|
Chris@13
|
402 cerr << "corresponding Hz " << coarseHz << " scores " << coarseScore << endl;
|
Chris@4
|
403
|
Chris@4
|
404 f.values.clear();
|
Chris@13
|
405 for (auto v: coarseFeature) f.values.push_back(v);
|
Chris@13
|
406 fs[m_outputs["rotfeature"]].push_back(f);
|
Chris@4
|
407
|
Chris@1
|
408 return fs;
|
Chris@0
|
409 }
|
Chris@0
|
410
|