Chris@152
|
1 /* -*- c-basic-offset: 4 indent-tabs-mode: nil -*- vi:set ts=8 sts=4 sw=4: */
|
Chris@152
|
2
|
Chris@152
|
3 /*
|
Chris@152
|
4 Sonic Visualiser
|
Chris@152
|
5 An audio file viewer and annotation editor.
|
Chris@152
|
6 Centre for Digital Music, Queen Mary, University of London.
|
Chris@152
|
7 This file copyright 2006 Chris Cannam.
|
Chris@152
|
8
|
Chris@152
|
9 This program is free software; you can redistribute it and/or
|
Chris@152
|
10 modify it under the terms of the GNU General Public License as
|
Chris@152
|
11 published by the Free Software Foundation; either version 2 of the
|
Chris@152
|
12 License, or (at your option) any later version. See the file
|
Chris@152
|
13 COPYING included with this distribution for more information.
|
Chris@152
|
14 */
|
Chris@152
|
15
|
Chris@152
|
16 #include "FFTModel.h"
|
Chris@152
|
17 #include "DenseTimeValueModel.h"
|
Chris@152
|
18
|
Chris@183
|
19 #include "base/Profiler.h"
|
Chris@275
|
20 #include "base/Pitch.h"
|
Chris@1256
|
21 #include "base/HitCount.h"
|
Chris@183
|
22
|
Chris@402
|
23 #include <algorithm>
|
Chris@402
|
24
|
Chris@152
|
25 #include <cassert>
|
Chris@1090
|
26 #include <deque>
|
Chris@152
|
27
|
Chris@1090
|
28 using namespace std;
|
Chris@1090
|
29
|
Chris@1256
|
30 static HitCount inSmallCache("FFTModel: Small FFT cache");
|
Chris@1256
|
31 static HitCount inSourceCache("FFTModel: Source data cache");
|
Chris@1256
|
32
|
Chris@152
|
33 FFTModel::FFTModel(const DenseTimeValueModel *model,
|
Chris@152
|
34 int channel,
|
Chris@152
|
35 WindowType windowType,
|
Chris@929
|
36 int windowSize,
|
Chris@929
|
37 int windowIncrement,
|
Chris@1090
|
38 int fftSize) :
|
Chris@1090
|
39 m_model(model),
|
Chris@1090
|
40 m_channel(channel),
|
Chris@1090
|
41 m_windowType(windowType),
|
Chris@1090
|
42 m_windowSize(windowSize),
|
Chris@1090
|
43 m_windowIncrement(windowIncrement),
|
Chris@1090
|
44 m_fftSize(fftSize),
|
Chris@1091
|
45 m_windower(windowType, windowSize),
|
Chris@1093
|
46 m_fft(fftSize),
|
Chris@1093
|
47 m_cacheSize(3)
|
Chris@152
|
48 {
|
Chris@1091
|
49 if (m_windowSize > m_fftSize) {
|
Chris@1091
|
50 cerr << "ERROR: FFTModel::FFTModel: window size (" << m_windowSize
|
Chris@1091
|
51 << ") must be at least FFT size (" << m_fftSize << ")" << endl;
|
Chris@1091
|
52 throw invalid_argument("FFTModel window size must be at least FFT size");
|
Chris@1091
|
53 }
|
Chris@1133
|
54
|
Chris@1270
|
55 m_fft.initFloat();
|
Chris@1270
|
56
|
Chris@1133
|
57 connect(model, SIGNAL(modelChanged()), this, SIGNAL(modelChanged()));
|
Chris@1133
|
58 connect(model, SIGNAL(modelChangedWithin(sv_frame_t, sv_frame_t)),
|
Chris@1133
|
59 this, SIGNAL(modelChangedWithin(sv_frame_t, sv_frame_t)));
|
Chris@152
|
60 }
|
Chris@152
|
61
|
Chris@152
|
62 FFTModel::~FFTModel()
|
Chris@152
|
63 {
|
Chris@152
|
64 }
|
Chris@152
|
65
|
Chris@360
|
66 void
|
Chris@360
|
67 FFTModel::sourceModelAboutToBeDeleted()
|
Chris@360
|
68 {
|
Chris@1090
|
69 if (m_model) {
|
Chris@1090
|
70 cerr << "FFTModel[" << this << "]::sourceModelAboutToBeDeleted(" << m_model << ")" << endl;
|
Chris@1090
|
71 m_model = 0;
|
Chris@360
|
72 }
|
Chris@360
|
73 }
|
Chris@360
|
74
|
Chris@1091
|
75 int
|
Chris@1091
|
76 FFTModel::getWidth() const
|
Chris@1091
|
77 {
|
Chris@1091
|
78 if (!m_model) return 0;
|
Chris@1091
|
79 return int((m_model->getEndFrame() - m_model->getStartFrame())
|
Chris@1091
|
80 / m_windowIncrement) + 1;
|
Chris@1091
|
81 }
|
Chris@1091
|
82
|
Chris@1091
|
83 int
|
Chris@1091
|
84 FFTModel::getHeight() const
|
Chris@1091
|
85 {
|
Chris@1091
|
86 return m_fftSize / 2 + 1;
|
Chris@1091
|
87 }
|
Chris@1091
|
88
|
Chris@152
|
89 QString
|
Chris@929
|
90 FFTModel::getBinName(int n) const
|
Chris@152
|
91 {
|
Chris@1040
|
92 sv_samplerate_t sr = getSampleRate();
|
Chris@152
|
93 if (!sr) return "";
|
Chris@204
|
94 QString name = tr("%1 Hz").arg((n * sr) / ((getHeight()-1) * 2));
|
Chris@152
|
95 return name;
|
Chris@152
|
96 }
|
Chris@152
|
97
|
Chris@1091
|
98 FFTModel::Column
|
Chris@1091
|
99 FFTModel::getColumn(int x) const
|
Chris@1091
|
100 {
|
Chris@1091
|
101 auto cplx = getFFTColumn(x);
|
Chris@1091
|
102 Column col;
|
Chris@1154
|
103 col.reserve(cplx.size());
|
Chris@1091
|
104 for (auto c: cplx) col.push_back(abs(c));
|
Chris@1154
|
105 return move(col);
|
Chris@1091
|
106 }
|
Chris@1091
|
107
|
Chris@1200
|
108 FFTModel::Column
|
Chris@1200
|
109 FFTModel::getPhases(int x) const
|
Chris@1200
|
110 {
|
Chris@1200
|
111 auto cplx = getFFTColumn(x);
|
Chris@1200
|
112 Column col;
|
Chris@1200
|
113 col.reserve(cplx.size());
|
Chris@1201
|
114 for (auto c: cplx) {
|
Chris@1201
|
115 col.push_back(arg(c));
|
Chris@1201
|
116 }
|
Chris@1200
|
117 return move(col);
|
Chris@1200
|
118 }
|
Chris@1200
|
119
|
Chris@1091
|
120 float
|
Chris@1091
|
121 FFTModel::getMagnitudeAt(int x, int y) const
|
Chris@1091
|
122 {
|
Chris@1093
|
123 if (x < 0 || x >= getWidth() || y < 0 || y >= getHeight()) return 0.f;
|
Chris@1093
|
124 auto col = getFFTColumn(x);
|
Chris@1093
|
125 return abs(col[y]);
|
Chris@1091
|
126 }
|
Chris@1091
|
127
|
Chris@1091
|
128 float
|
Chris@1091
|
129 FFTModel::getMaximumMagnitudeAt(int x) const
|
Chris@1091
|
130 {
|
Chris@1091
|
131 Column col(getColumn(x));
|
Chris@1092
|
132 float max = 0.f;
|
Chris@1154
|
133 int n = int(col.size());
|
Chris@1154
|
134 for (int i = 0; i < n; ++i) {
|
Chris@1092
|
135 if (col[i] > max) max = col[i];
|
Chris@1092
|
136 }
|
Chris@1092
|
137 return max;
|
Chris@1091
|
138 }
|
Chris@1091
|
139
|
Chris@1091
|
140 float
|
Chris@1091
|
141 FFTModel::getPhaseAt(int x, int y) const
|
Chris@1091
|
142 {
|
Chris@1093
|
143 if (x < 0 || x >= getWidth() || y < 0 || y >= getHeight()) return 0.f;
|
Chris@1091
|
144 return arg(getFFTColumn(x)[y]);
|
Chris@1091
|
145 }
|
Chris@1091
|
146
|
Chris@1091
|
147 void
|
Chris@1091
|
148 FFTModel::getValuesAt(int x, int y, float &re, float &im) const
|
Chris@1091
|
149 {
|
Chris@1091
|
150 auto col = getFFTColumn(x);
|
Chris@1091
|
151 re = col[y].real();
|
Chris@1091
|
152 im = col[y].imag();
|
Chris@1091
|
153 }
|
Chris@1091
|
154
|
Chris@1091
|
155 bool
|
Chris@1091
|
156 FFTModel::getMagnitudesAt(int x, float *values, int minbin, int count) const
|
Chris@1091
|
157 {
|
Chris@1091
|
158 if (count == 0) count = getHeight();
|
Chris@1091
|
159 auto col = getFFTColumn(x);
|
Chris@1091
|
160 for (int i = 0; i < count; ++i) {
|
Chris@1091
|
161 values[i] = abs(col[minbin + i]);
|
Chris@1091
|
162 }
|
Chris@1091
|
163 return true;
|
Chris@1091
|
164 }
|
Chris@1091
|
165
|
Chris@1091
|
166 bool
|
Chris@1091
|
167 FFTModel::getPhasesAt(int x, float *values, int minbin, int count) const
|
Chris@1091
|
168 {
|
Chris@1091
|
169 if (count == 0) count = getHeight();
|
Chris@1091
|
170 auto col = getFFTColumn(x);
|
Chris@1091
|
171 for (int i = 0; i < count; ++i) {
|
Chris@1091
|
172 values[i] = arg(col[minbin + i]);
|
Chris@1091
|
173 }
|
Chris@1091
|
174 return true;
|
Chris@1091
|
175 }
|
Chris@1091
|
176
|
Chris@1091
|
177 bool
|
Chris@1091
|
178 FFTModel::getValuesAt(int x, float *reals, float *imags, int minbin, int count) const
|
Chris@1091
|
179 {
|
Chris@1091
|
180 if (count == 0) count = getHeight();
|
Chris@1091
|
181 auto col = getFFTColumn(x);
|
Chris@1091
|
182 for (int i = 0; i < count; ++i) {
|
Chris@1091
|
183 reals[i] = col[minbin + i].real();
|
Chris@1091
|
184 }
|
Chris@1091
|
185 for (int i = 0; i < count; ++i) {
|
Chris@1091
|
186 imags[i] = col[minbin + i].imag();
|
Chris@1091
|
187 }
|
Chris@1091
|
188 return true;
|
Chris@1091
|
189 }
|
Chris@1091
|
190
|
Chris@1091
|
191 vector<float>
|
Chris@1091
|
192 FFTModel::getSourceSamples(int column) const
|
Chris@1091
|
193 {
|
Chris@1094
|
194 // m_fftSize may be greater than m_windowSize, but not the reverse
|
Chris@1094
|
195
|
Chris@1094
|
196 // cerr << "getSourceSamples(" << column << ")" << endl;
|
Chris@1094
|
197
|
Chris@1091
|
198 auto range = getSourceSampleRange(column);
|
Chris@1094
|
199 auto data = getSourceData(range);
|
Chris@1094
|
200
|
Chris@1091
|
201 int off = (m_fftSize - m_windowSize) / 2;
|
Chris@1094
|
202
|
Chris@1094
|
203 if (off == 0) {
|
Chris@1094
|
204 return data;
|
Chris@1094
|
205 } else {
|
Chris@1094
|
206 vector<float> pad(off, 0.f);
|
Chris@1094
|
207 vector<float> padded;
|
Chris@1094
|
208 padded.reserve(m_fftSize);
|
Chris@1094
|
209 padded.insert(padded.end(), pad.begin(), pad.end());
|
Chris@1094
|
210 padded.insert(padded.end(), data.begin(), data.end());
|
Chris@1094
|
211 padded.insert(padded.end(), pad.begin(), pad.end());
|
Chris@1094
|
212 return padded;
|
Chris@1094
|
213 }
|
Chris@1094
|
214 }
|
Chris@1094
|
215
|
Chris@1094
|
216 vector<float>
|
Chris@1094
|
217 FFTModel::getSourceData(pair<sv_frame_t, sv_frame_t> range) const
|
Chris@1094
|
218 {
|
Chris@1094
|
219 // cerr << "getSourceData(" << range.first << "," << range.second
|
Chris@1094
|
220 // << "): saved range is (" << m_savedData.range.first
|
Chris@1094
|
221 // << "," << m_savedData.range.second << ")" << endl;
|
Chris@1094
|
222
|
Chris@1100
|
223 if (m_savedData.range == range) {
|
Chris@1256
|
224 inSourceCache.hit();
|
Chris@1100
|
225 return m_savedData.data;
|
Chris@1100
|
226 }
|
Chris@1094
|
227
|
Chris@1270
|
228 Profiler profiler("FFTModel::getSourceData (cache miss)");
|
Chris@1270
|
229
|
Chris@1094
|
230 if (range.first < m_savedData.range.second &&
|
Chris@1094
|
231 range.first >= m_savedData.range.first &&
|
Chris@1094
|
232 range.second > m_savedData.range.second) {
|
Chris@1094
|
233
|
Chris@1256
|
234 inSourceCache.partial();
|
Chris@1256
|
235
|
Chris@1100
|
236 sv_frame_t discard = range.first - m_savedData.range.first;
|
Chris@1100
|
237
|
Chris@1100
|
238 vector<float> acc(m_savedData.data.begin() + discard,
|
Chris@1100
|
239 m_savedData.data.end());
|
Chris@1094
|
240
|
Chris@1095
|
241 vector<float> rest =
|
Chris@1095
|
242 getSourceDataUncached({ m_savedData.range.second, range.second });
|
Chris@1100
|
243
|
Chris@1100
|
244 acc.insert(acc.end(), rest.begin(), rest.end());
|
Chris@1094
|
245
|
Chris@1095
|
246 m_savedData = { range, acc };
|
Chris@1095
|
247 return acc;
|
Chris@1095
|
248
|
Chris@1095
|
249 } else {
|
Chris@1095
|
250
|
Chris@1256
|
251 inSourceCache.miss();
|
Chris@1256
|
252
|
Chris@1095
|
253 auto data = getSourceDataUncached(range);
|
Chris@1095
|
254 m_savedData = { range, data };
|
Chris@1095
|
255 return data;
|
Chris@1094
|
256 }
|
Chris@1095
|
257 }
|
Chris@1094
|
258
|
Chris@1095
|
259 vector<float>
|
Chris@1095
|
260 FFTModel::getSourceDataUncached(pair<sv_frame_t, sv_frame_t> range) const
|
Chris@1095
|
261 {
|
Chris@1091
|
262 decltype(range.first) pfx = 0;
|
Chris@1091
|
263 if (range.first < 0) {
|
Chris@1091
|
264 pfx = -range.first;
|
Chris@1091
|
265 range = { 0, range.second };
|
Chris@1091
|
266 }
|
Chris@1096
|
267
|
Chris@1096
|
268 auto data = m_model->getData(m_channel,
|
Chris@1096
|
269 range.first,
|
Chris@1096
|
270 range.second - range.first);
|
Chris@1096
|
271
|
Chris@1096
|
272 // don't return a partial frame
|
Chris@1096
|
273 data.resize(range.second - range.first, 0.f);
|
Chris@1096
|
274
|
Chris@1096
|
275 if (pfx > 0) {
|
Chris@1096
|
276 vector<float> pad(pfx, 0.f);
|
Chris@1096
|
277 data.insert(data.begin(), pad.begin(), pad.end());
|
Chris@1096
|
278 }
|
Chris@1096
|
279
|
Chris@1091
|
280 if (m_channel == -1) {
|
Chris@1091
|
281 int channels = m_model->getChannelCount();
|
Chris@1091
|
282 if (channels > 1) {
|
Chris@1096
|
283 int n = int(data.size());
|
Chris@1096
|
284 float factor = 1.f / float(channels);
|
Chris@1100
|
285 // use mean instead of sum for fft model input
|
Chris@1096
|
286 for (int i = 0; i < n; ++i) {
|
Chris@1096
|
287 data[i] *= factor;
|
Chris@1091
|
288 }
|
Chris@1091
|
289 }
|
Chris@1091
|
290 }
|
Chris@1094
|
291
|
Chris@1094
|
292 return data;
|
Chris@1091
|
293 }
|
Chris@1091
|
294
|
Chris@1091
|
295 vector<complex<float>>
|
Chris@1093
|
296 FFTModel::getFFTColumn(int n) const
|
Chris@1091
|
297 {
|
Chris@1258
|
298 // The small cache (i.e. the m_cached deque) is for cases where
|
Chris@1258
|
299 // values are looked up individually, and for e.g. peak-frequency
|
Chris@1258
|
300 // spectrograms where values from two consecutive columns are
|
Chris@1257
|
301 // needed at once. This cache gets essentially no hits when
|
Chris@1258
|
302 // scrolling through a magnitude spectrogram, but 95%+ hits with a
|
Chris@1258
|
303 // peak-frequency spectrogram.
|
Chris@1257
|
304 for (const auto &incache : m_cached) {
|
Chris@1093
|
305 if (incache.n == n) {
|
Chris@1256
|
306 inSmallCache.hit();
|
Chris@1093
|
307 return incache.col;
|
Chris@1093
|
308 }
|
Chris@1093
|
309 }
|
Chris@1256
|
310 inSmallCache.miss();
|
Chris@1258
|
311
|
Chris@1258
|
312 Profiler profiler("FFTModel::getFFTColumn (cache miss)");
|
Chris@1093
|
313
|
Chris@1093
|
314 auto samples = getSourceSamples(n);
|
Chris@1100
|
315 m_windower.cut(samples.data());
|
Chris@1270
|
316 breakfastquay::v_fftshift(samples.data(), m_fftSize);
|
Chris@1270
|
317
|
Chris@1270
|
318 vector<complex<float>> col(m_fftSize/2 + 1);
|
Chris@1270
|
319
|
Chris@1270
|
320 m_fft.forwardInterleaved(samples.data(),
|
Chris@1270
|
321 reinterpret_cast<float *>(col.data()));
|
Chris@1093
|
322
|
Chris@1093
|
323 SavedColumn sc { n, col };
|
Chris@1093
|
324 if (m_cached.size() >= m_cacheSize) {
|
Chris@1093
|
325 m_cached.pop_front();
|
Chris@1093
|
326 }
|
Chris@1093
|
327 m_cached.push_back(sc);
|
Chris@1093
|
328
|
Chris@1154
|
329 return move(col);
|
Chris@1091
|
330 }
|
Chris@1091
|
331
|
Chris@275
|
332 bool
|
Chris@1045
|
333 FFTModel::estimateStableFrequency(int x, int y, double &frequency)
|
Chris@275
|
334 {
|
Chris@275
|
335 if (!isOK()) return false;
|
Chris@275
|
336
|
Chris@1090
|
337 frequency = double(y * getSampleRate()) / m_fftSize;
|
Chris@275
|
338
|
Chris@275
|
339 if (x+1 >= getWidth()) return false;
|
Chris@275
|
340
|
Chris@275
|
341 // At frequency f, a phase shift of 2pi (one cycle) happens in 1/f sec.
|
Chris@275
|
342 // At hopsize h and sample rate sr, one hop happens in h/sr sec.
|
Chris@275
|
343 // At window size w, for bin b, f is b*sr/w.
|
Chris@275
|
344 // thus 2pi phase shift happens in w/(b*sr) sec.
|
Chris@275
|
345 // We need to know what phase shift we expect from h/sr sec.
|
Chris@275
|
346 // -> 2pi * ((h/sr) / (w/(b*sr)))
|
Chris@275
|
347 // = 2pi * ((h * b * sr) / (w * sr))
|
Chris@275
|
348 // = 2pi * (h * b) / w.
|
Chris@275
|
349
|
Chris@1038
|
350 double oldPhase = getPhaseAt(x, y);
|
Chris@1038
|
351 double newPhase = getPhaseAt(x+1, y);
|
Chris@275
|
352
|
Chris@929
|
353 int incr = getResolution();
|
Chris@275
|
354
|
Chris@1090
|
355 double expectedPhase = oldPhase + (2.0 * M_PI * y * incr) / m_fftSize;
|
Chris@275
|
356
|
Chris@1038
|
357 double phaseError = princarg(newPhase - expectedPhase);
|
Chris@275
|
358
|
Chris@275
|
359 // The new frequency estimate based on the phase error resulting
|
Chris@275
|
360 // from assuming the "native" frequency of this bin
|
Chris@275
|
361
|
Chris@275
|
362 frequency =
|
Chris@1090
|
363 (getSampleRate() * (expectedPhase + phaseError - oldPhase)) /
|
Chris@1045
|
364 (2.0 * M_PI * incr);
|
Chris@275
|
365
|
Chris@275
|
366 return true;
|
Chris@275
|
367 }
|
Chris@275
|
368
|
Chris@275
|
369 FFTModel::PeakLocationSet
|
Chris@1191
|
370 FFTModel::getPeaks(PeakPickType type, int x, int ymin, int ymax) const
|
Chris@275
|
371 {
|
Chris@551
|
372 Profiler profiler("FFTModel::getPeaks");
|
Chris@551
|
373
|
Chris@275
|
374 FFTModel::PeakLocationSet peaks;
|
Chris@275
|
375 if (!isOK()) return peaks;
|
Chris@275
|
376
|
Chris@275
|
377 if (ymax == 0 || ymax > getHeight() - 1) {
|
Chris@275
|
378 ymax = getHeight() - 1;
|
Chris@275
|
379 }
|
Chris@275
|
380
|
Chris@275
|
381 if (type == AllPeaks) {
|
Chris@551
|
382 int minbin = ymin;
|
Chris@551
|
383 if (minbin > 0) minbin = minbin - 1;
|
Chris@551
|
384 int maxbin = ymax;
|
Chris@551
|
385 if (maxbin < getHeight() - 1) maxbin = maxbin + 1;
|
Chris@551
|
386 const int n = maxbin - minbin + 1;
|
Chris@1218
|
387 float *values = new float[n];
|
Chris@551
|
388 getMagnitudesAt(x, values, minbin, maxbin - minbin + 1);
|
Chris@929
|
389 for (int bin = ymin; bin <= ymax; ++bin) {
|
Chris@551
|
390 if (bin == minbin || bin == maxbin) continue;
|
Chris@551
|
391 if (values[bin - minbin] > values[bin - minbin - 1] &&
|
Chris@551
|
392 values[bin - minbin] > values[bin - minbin + 1]) {
|
Chris@275
|
393 peaks.insert(bin);
|
Chris@275
|
394 }
|
Chris@275
|
395 }
|
Chris@1218
|
396 delete[] values;
|
Chris@275
|
397 return peaks;
|
Chris@275
|
398 }
|
Chris@275
|
399
|
Chris@551
|
400 Column values = getColumn(x);
|
Chris@1154
|
401 int nv = int(values.size());
|
Chris@275
|
402
|
Chris@500
|
403 float mean = 0.f;
|
Chris@1154
|
404 for (int i = 0; i < nv; ++i) mean += values[i];
|
Chris@1154
|
405 if (nv > 0) mean = mean / float(values.size());
|
Chris@1038
|
406
|
Chris@275
|
407 // For peak picking we use a moving median window, picking the
|
Chris@275
|
408 // highest value within each continuous region of values that
|
Chris@275
|
409 // exceed the median. For pitch adaptivity, we adjust the window
|
Chris@275
|
410 // size to a roughly constant pitch range (about four tones).
|
Chris@275
|
411
|
Chris@1040
|
412 sv_samplerate_t sampleRate = getSampleRate();
|
Chris@275
|
413
|
Chris@1090
|
414 deque<float> window;
|
Chris@1090
|
415 vector<int> inrange;
|
Chris@280
|
416 float dist = 0.5;
|
Chris@500
|
417
|
Chris@929
|
418 int medianWinSize = getPeakPickWindowSize(type, sampleRate, ymin, dist);
|
Chris@929
|
419 int halfWin = medianWinSize/2;
|
Chris@275
|
420
|
Chris@929
|
421 int binmin;
|
Chris@275
|
422 if (ymin > halfWin) binmin = ymin - halfWin;
|
Chris@275
|
423 else binmin = 0;
|
Chris@275
|
424
|
Chris@929
|
425 int binmax;
|
Chris@1154
|
426 if (ymax + halfWin < nv) binmax = ymax + halfWin;
|
Chris@1154
|
427 else binmax = nv - 1;
|
Chris@275
|
428
|
Chris@929
|
429 int prevcentre = 0;
|
Chris@500
|
430
|
Chris@929
|
431 for (int bin = binmin; bin <= binmax; ++bin) {
|
Chris@275
|
432
|
Chris@275
|
433 float value = values[bin];
|
Chris@275
|
434
|
Chris@275
|
435 window.push_back(value);
|
Chris@275
|
436
|
Chris@280
|
437 // so-called median will actually be the dist*100'th percentile
|
Chris@280
|
438 medianWinSize = getPeakPickWindowSize(type, sampleRate, bin, dist);
|
Chris@275
|
439 halfWin = medianWinSize/2;
|
Chris@275
|
440
|
Chris@929
|
441 while ((int)window.size() > medianWinSize) {
|
Chris@500
|
442 window.pop_front();
|
Chris@500
|
443 }
|
Chris@500
|
444
|
Chris@1038
|
445 int actualSize = int(window.size());
|
Chris@275
|
446
|
Chris@275
|
447 if (type == MajorPitchAdaptivePeaks) {
|
Chris@1154
|
448 if (ymax + halfWin < nv) binmax = ymax + halfWin;
|
Chris@1154
|
449 else binmax = nv - 1;
|
Chris@275
|
450 }
|
Chris@275
|
451
|
Chris@1090
|
452 deque<float> sorted(window);
|
Chris@1090
|
453 sort(sorted.begin(), sorted.end());
|
Chris@1038
|
454 float median = sorted[int(float(sorted.size()) * dist)];
|
Chris@275
|
455
|
Chris@929
|
456 int centrebin = 0;
|
Chris@500
|
457 if (bin > actualSize/2) centrebin = bin - actualSize/2;
|
Chris@500
|
458
|
Chris@500
|
459 while (centrebin > prevcentre || bin == binmin) {
|
Chris@275
|
460
|
Chris@500
|
461 if (centrebin > prevcentre) ++prevcentre;
|
Chris@500
|
462
|
Chris@500
|
463 float centre = values[prevcentre];
|
Chris@500
|
464
|
Chris@500
|
465 if (centre > median) {
|
Chris@500
|
466 inrange.push_back(centrebin);
|
Chris@500
|
467 }
|
Chris@500
|
468
|
Chris@1154
|
469 if (centre <= median || centrebin+1 == nv) {
|
Chris@500
|
470 if (!inrange.empty()) {
|
Chris@929
|
471 int peakbin = 0;
|
Chris@500
|
472 float peakval = 0.f;
|
Chris@929
|
473 for (int i = 0; i < (int)inrange.size(); ++i) {
|
Chris@500
|
474 if (i == 0 || values[inrange[i]] > peakval) {
|
Chris@500
|
475 peakval = values[inrange[i]];
|
Chris@500
|
476 peakbin = inrange[i];
|
Chris@500
|
477 }
|
Chris@500
|
478 }
|
Chris@500
|
479 inrange.clear();
|
Chris@500
|
480 if (peakbin >= ymin && peakbin <= ymax) {
|
Chris@500
|
481 peaks.insert(peakbin);
|
Chris@275
|
482 }
|
Chris@275
|
483 }
|
Chris@275
|
484 }
|
Chris@500
|
485
|
Chris@500
|
486 if (bin == binmin) break;
|
Chris@275
|
487 }
|
Chris@275
|
488 }
|
Chris@275
|
489
|
Chris@275
|
490 return peaks;
|
Chris@275
|
491 }
|
Chris@275
|
492
|
Chris@929
|
493 int
|
Chris@1040
|
494 FFTModel::getPeakPickWindowSize(PeakPickType type, sv_samplerate_t sampleRate,
|
Chris@929
|
495 int bin, float &percentile) const
|
Chris@275
|
496 {
|
Chris@280
|
497 percentile = 0.5;
|
Chris@275
|
498 if (type == MajorPeaks) return 10;
|
Chris@275
|
499 if (bin == 0) return 3;
|
Chris@280
|
500
|
Chris@1091
|
501 double binfreq = (sampleRate * bin) / m_fftSize;
|
Chris@1038
|
502 double hifreq = Pitch::getFrequencyForPitch(73, 0, binfreq);
|
Chris@280
|
503
|
Chris@1091
|
504 int hibin = int(lrint((hifreq * m_fftSize) / sampleRate));
|
Chris@275
|
505 int medianWinSize = hibin - bin;
|
Chris@275
|
506 if (medianWinSize < 3) medianWinSize = 3;
|
Chris@280
|
507
|
Chris@1091
|
508 percentile = 0.5f + float(binfreq / sampleRate);
|
Chris@280
|
509
|
Chris@275
|
510 return medianWinSize;
|
Chris@275
|
511 }
|
Chris@275
|
512
|
Chris@275
|
513 FFTModel::PeakSet
|
Chris@929
|
514 FFTModel::getPeakFrequencies(PeakPickType type, int x,
|
Chris@1191
|
515 int ymin, int ymax) const
|
Chris@275
|
516 {
|
Chris@551
|
517 Profiler profiler("FFTModel::getPeakFrequencies");
|
Chris@551
|
518
|
Chris@275
|
519 PeakSet peaks;
|
Chris@275
|
520 if (!isOK()) return peaks;
|
Chris@275
|
521 PeakLocationSet locations = getPeaks(type, x, ymin, ymax);
|
Chris@275
|
522
|
Chris@1040
|
523 sv_samplerate_t sampleRate = getSampleRate();
|
Chris@929
|
524 int incr = getResolution();
|
Chris@275
|
525
|
Chris@275
|
526 // This duplicates some of the work of estimateStableFrequency to
|
Chris@275
|
527 // allow us to retrieve the phases in two separate vertical
|
Chris@275
|
528 // columns, instead of jumping back and forth between columns x and
|
Chris@275
|
529 // x+1, which may be significantly slower if re-seeking is needed
|
Chris@275
|
530
|
Chris@1090
|
531 vector<float> phases;
|
Chris@275
|
532 for (PeakLocationSet::iterator i = locations.begin();
|
Chris@275
|
533 i != locations.end(); ++i) {
|
Chris@275
|
534 phases.push_back(getPhaseAt(x, *i));
|
Chris@275
|
535 }
|
Chris@275
|
536
|
Chris@929
|
537 int phaseIndex = 0;
|
Chris@275
|
538 for (PeakLocationSet::iterator i = locations.begin();
|
Chris@275
|
539 i != locations.end(); ++i) {
|
Chris@1038
|
540 double oldPhase = phases[phaseIndex];
|
Chris@1038
|
541 double newPhase = getPhaseAt(x+1, *i);
|
Chris@1090
|
542 double expectedPhase = oldPhase + (2.0 * M_PI * *i * incr) / m_fftSize;
|
Chris@1038
|
543 double phaseError = princarg(newPhase - expectedPhase);
|
Chris@1038
|
544 double frequency =
|
Chris@275
|
545 (sampleRate * (expectedPhase + phaseError - oldPhase))
|
Chris@275
|
546 / (2 * M_PI * incr);
|
Chris@1045
|
547 peaks[*i] = frequency;
|
Chris@275
|
548 ++phaseIndex;
|
Chris@275
|
549 }
|
Chris@275
|
550
|
Chris@275
|
551 return peaks;
|
Chris@275
|
552 }
|
Chris@275
|
553
|