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@297
|
18 #include "AggregateWaveModel.h"
|
Chris@152
|
19
|
Chris@183
|
20 #include "base/Profiler.h"
|
Chris@275
|
21 #include "base/Pitch.h"
|
Chris@183
|
22
|
Chris@152
|
23 #include <cassert>
|
Chris@152
|
24
|
Chris@152
|
25 FFTModel::FFTModel(const DenseTimeValueModel *model,
|
Chris@152
|
26 int channel,
|
Chris@152
|
27 WindowType windowType,
|
Chris@152
|
28 size_t windowSize,
|
Chris@152
|
29 size_t windowIncrement,
|
Chris@152
|
30 size_t fftSize,
|
Chris@152
|
31 bool polar,
|
Chris@334
|
32 StorageAdviser::Criteria criteria,
|
Chris@152
|
33 size_t fillFromColumn) :
|
Chris@152
|
34 //!!! ZoomConstraint!
|
Chris@152
|
35 m_server(0),
|
Chris@152
|
36 m_xshift(0),
|
Chris@152
|
37 m_yshift(0)
|
Chris@152
|
38 {
|
Chris@297
|
39 setSourceModel(const_cast<DenseTimeValueModel *>(model)); //!!! hmm.
|
Chris@297
|
40
|
Chris@297
|
41 m_server = getServer(model,
|
Chris@297
|
42 channel,
|
Chris@297
|
43 windowType,
|
Chris@297
|
44 windowSize,
|
Chris@297
|
45 windowIncrement,
|
Chris@297
|
46 fftSize,
|
Chris@297
|
47 polar,
|
Chris@334
|
48 criteria,
|
Chris@297
|
49 fillFromColumn);
|
Chris@152
|
50
|
Chris@200
|
51 if (!m_server) return; // caller should check isOK()
|
Chris@200
|
52
|
Chris@152
|
53 size_t xratio = windowIncrement / m_server->getWindowIncrement();
|
Chris@152
|
54 size_t yratio = m_server->getFFTSize() / fftSize;
|
Chris@152
|
55
|
Chris@152
|
56 while (xratio > 1) {
|
Chris@152
|
57 if (xratio & 0x1) {
|
Chris@152
|
58 std::cerr << "ERROR: FFTModel: Window increment ratio "
|
Chris@152
|
59 << windowIncrement << " / "
|
Chris@152
|
60 << m_server->getWindowIncrement()
|
Chris@152
|
61 << " must be a power of two" << std::endl;
|
Chris@152
|
62 assert(!(xratio & 0x1));
|
Chris@152
|
63 }
|
Chris@152
|
64 ++m_xshift;
|
Chris@152
|
65 xratio >>= 1;
|
Chris@152
|
66 }
|
Chris@152
|
67
|
Chris@152
|
68 while (yratio > 1) {
|
Chris@152
|
69 if (yratio & 0x1) {
|
Chris@152
|
70 std::cerr << "ERROR: FFTModel: FFT size ratio "
|
Chris@152
|
71 << m_server->getFFTSize() << " / " << fftSize
|
Chris@152
|
72 << " must be a power of two" << std::endl;
|
Chris@152
|
73 assert(!(yratio & 0x1));
|
Chris@152
|
74 }
|
Chris@152
|
75 ++m_yshift;
|
Chris@152
|
76 yratio >>= 1;
|
Chris@152
|
77 }
|
Chris@152
|
78 }
|
Chris@152
|
79
|
Chris@152
|
80 FFTModel::~FFTModel()
|
Chris@152
|
81 {
|
Chris@200
|
82 if (m_server) FFTDataServer::releaseInstance(m_server);
|
Chris@152
|
83 }
|
Chris@152
|
84
|
Chris@297
|
85 FFTDataServer *
|
Chris@297
|
86 FFTModel::getServer(const DenseTimeValueModel *model,
|
Chris@297
|
87 int channel,
|
Chris@297
|
88 WindowType windowType,
|
Chris@297
|
89 size_t windowSize,
|
Chris@297
|
90 size_t windowIncrement,
|
Chris@297
|
91 size_t fftSize,
|
Chris@297
|
92 bool polar,
|
Chris@334
|
93 StorageAdviser::Criteria criteria,
|
Chris@297
|
94 size_t fillFromColumn)
|
Chris@297
|
95 {
|
Chris@297
|
96 // Obviously, an FFT model of channel C (where C != -1) of an
|
Chris@297
|
97 // aggregate model is the same as the FFT model of the appropriate
|
Chris@297
|
98 // channel of whichever model that aggregate channel is drawn
|
Chris@297
|
99 // from. We should use that model here, in case we already have
|
Chris@297
|
100 // the data for it or will be wanting the same data again later.
|
Chris@297
|
101
|
Chris@297
|
102 // If the channel is -1 (i.e. mixture of all channels), then we
|
Chris@297
|
103 // can't do this shortcut unless the aggregate model only has one
|
Chris@297
|
104 // channel or contains exactly all of the channels of a single
|
Chris@297
|
105 // other model. That isn't very likely -- if it were the case,
|
Chris@297
|
106 // why would we be using an aggregate model?
|
Chris@297
|
107
|
Chris@297
|
108 if (channel >= 0) {
|
Chris@297
|
109
|
Chris@297
|
110 const AggregateWaveModel *aggregate =
|
Chris@297
|
111 dynamic_cast<const AggregateWaveModel *>(model);
|
Chris@297
|
112
|
Chris@297
|
113 if (aggregate && channel < aggregate->getComponentCount()) {
|
Chris@297
|
114
|
Chris@297
|
115 AggregateWaveModel::ModelChannelSpec spec =
|
Chris@297
|
116 aggregate->getComponent(channel);
|
Chris@297
|
117
|
Chris@297
|
118 return getServer(spec.model,
|
Chris@297
|
119 spec.channel,
|
Chris@297
|
120 windowType,
|
Chris@297
|
121 windowSize,
|
Chris@297
|
122 windowIncrement,
|
Chris@297
|
123 fftSize,
|
Chris@297
|
124 polar,
|
Chris@334
|
125 criteria,
|
Chris@297
|
126 fillFromColumn);
|
Chris@297
|
127 }
|
Chris@297
|
128 }
|
Chris@297
|
129
|
Chris@297
|
130 // The normal case
|
Chris@297
|
131
|
Chris@297
|
132 return FFTDataServer::getFuzzyInstance(model,
|
Chris@297
|
133 channel,
|
Chris@297
|
134 windowType,
|
Chris@297
|
135 windowSize,
|
Chris@297
|
136 windowIncrement,
|
Chris@297
|
137 fftSize,
|
Chris@297
|
138 polar,
|
Chris@334
|
139 criteria,
|
Chris@297
|
140 fillFromColumn);
|
Chris@297
|
141 }
|
Chris@297
|
142
|
Chris@152
|
143 size_t
|
Chris@152
|
144 FFTModel::getSampleRate() const
|
Chris@152
|
145 {
|
Chris@152
|
146 return isOK() ? m_server->getModel()->getSampleRate() : 0;
|
Chris@152
|
147 }
|
Chris@152
|
148
|
Chris@152
|
149 void
|
Chris@182
|
150 FFTModel::getColumn(size_t x, Column &result) const
|
Chris@152
|
151 {
|
Chris@183
|
152 Profiler profiler("FFTModel::getColumn", false);
|
Chris@183
|
153
|
Chris@152
|
154 result.clear();
|
Chris@152
|
155 size_t height(getHeight());
|
Chris@152
|
156 for (size_t y = 0; y < height; ++y) {
|
Chris@152
|
157 result.push_back(const_cast<FFTModel *>(this)->getMagnitudeAt(x, y));
|
Chris@152
|
158 }
|
Chris@152
|
159 }
|
Chris@152
|
160
|
Chris@152
|
161 QString
|
Chris@152
|
162 FFTModel::getBinName(size_t n) const
|
Chris@152
|
163 {
|
Chris@152
|
164 size_t sr = getSampleRate();
|
Chris@152
|
165 if (!sr) return "";
|
Chris@204
|
166 QString name = tr("%1 Hz").arg((n * sr) / ((getHeight()-1) * 2));
|
Chris@152
|
167 return name;
|
Chris@152
|
168 }
|
Chris@152
|
169
|
Chris@275
|
170 bool
|
Chris@275
|
171 FFTModel::estimateStableFrequency(size_t x, size_t y, float &frequency)
|
Chris@275
|
172 {
|
Chris@275
|
173 if (!isOK()) return false;
|
Chris@275
|
174
|
Chris@275
|
175 size_t sampleRate = m_server->getModel()->getSampleRate();
|
Chris@275
|
176
|
Chris@275
|
177 size_t fftSize = m_server->getFFTSize() >> m_yshift;
|
Chris@275
|
178 frequency = (float(y) * sampleRate) / fftSize;
|
Chris@275
|
179
|
Chris@275
|
180 if (x+1 >= getWidth()) return false;
|
Chris@275
|
181
|
Chris@275
|
182 // At frequency f, a phase shift of 2pi (one cycle) happens in 1/f sec.
|
Chris@275
|
183 // At hopsize h and sample rate sr, one hop happens in h/sr sec.
|
Chris@275
|
184 // At window size w, for bin b, f is b*sr/w.
|
Chris@275
|
185 // thus 2pi phase shift happens in w/(b*sr) sec.
|
Chris@275
|
186 // We need to know what phase shift we expect from h/sr sec.
|
Chris@275
|
187 // -> 2pi * ((h/sr) / (w/(b*sr)))
|
Chris@275
|
188 // = 2pi * ((h * b * sr) / (w * sr))
|
Chris@275
|
189 // = 2pi * (h * b) / w.
|
Chris@275
|
190
|
Chris@275
|
191 float oldPhase = getPhaseAt(x, y);
|
Chris@275
|
192 float newPhase = getPhaseAt(x+1, y);
|
Chris@275
|
193
|
Chris@275
|
194 size_t incr = getResolution();
|
Chris@275
|
195
|
Chris@275
|
196 float expectedPhase = oldPhase + (2.0 * M_PI * y * incr) / fftSize;
|
Chris@275
|
197
|
Chris@275
|
198 float phaseError = princargf(newPhase - expectedPhase);
|
Chris@275
|
199
|
Chris@275
|
200 // bool stable = (fabsf(phaseError) < (1.1f * (m_windowIncrement * M_PI) / m_fftSize));
|
Chris@275
|
201
|
Chris@275
|
202 // The new frequency estimate based on the phase error resulting
|
Chris@275
|
203 // from assuming the "native" frequency of this bin
|
Chris@275
|
204
|
Chris@275
|
205 frequency =
|
Chris@275
|
206 (sampleRate * (expectedPhase + phaseError - oldPhase)) /
|
Chris@275
|
207 (2 * M_PI * incr);
|
Chris@275
|
208
|
Chris@275
|
209 return true;
|
Chris@275
|
210 }
|
Chris@275
|
211
|
Chris@275
|
212 FFTModel::PeakLocationSet
|
Chris@275
|
213 FFTModel::getPeaks(PeakPickType type, size_t x, size_t ymin, size_t ymax)
|
Chris@275
|
214 {
|
Chris@275
|
215 FFTModel::PeakLocationSet peaks;
|
Chris@275
|
216 if (!isOK()) return peaks;
|
Chris@275
|
217
|
Chris@275
|
218 if (ymax == 0 || ymax > getHeight() - 1) {
|
Chris@275
|
219 ymax = getHeight() - 1;
|
Chris@275
|
220 }
|
Chris@275
|
221
|
Chris@275
|
222 Column values;
|
Chris@275
|
223
|
Chris@275
|
224 if (type == AllPeaks) {
|
Chris@275
|
225 for (size_t y = ymin; y <= ymax; ++y) {
|
Chris@275
|
226 values.push_back(getMagnitudeAt(x, y));
|
Chris@275
|
227 }
|
Chris@275
|
228 size_t i = 0;
|
Chris@275
|
229 for (size_t bin = ymin; bin <= ymax; ++bin) {
|
Chris@275
|
230 if ((i == 0 || values[i] > values[i-1]) &&
|
Chris@275
|
231 (i == values.size()-1 || values[i] >= values[i+1])) {
|
Chris@275
|
232 peaks.insert(bin);
|
Chris@275
|
233 }
|
Chris@275
|
234 ++i;
|
Chris@275
|
235 }
|
Chris@275
|
236 return peaks;
|
Chris@275
|
237 }
|
Chris@275
|
238
|
Chris@275
|
239 getColumn(x, values);
|
Chris@275
|
240
|
Chris@275
|
241 // For peak picking we use a moving median window, picking the
|
Chris@275
|
242 // highest value within each continuous region of values that
|
Chris@275
|
243 // exceed the median. For pitch adaptivity, we adjust the window
|
Chris@275
|
244 // size to a roughly constant pitch range (about four tones).
|
Chris@275
|
245
|
Chris@275
|
246 size_t sampleRate = getSampleRate();
|
Chris@275
|
247
|
Chris@275
|
248 std::deque<float> window;
|
Chris@275
|
249 std::vector<size_t> inrange;
|
Chris@280
|
250 float dist = 0.5;
|
Chris@280
|
251 size_t medianWinSize = getPeakPickWindowSize(type, sampleRate, ymin, dist);
|
Chris@275
|
252 size_t halfWin = medianWinSize/2;
|
Chris@275
|
253
|
Chris@275
|
254 size_t binmin;
|
Chris@275
|
255 if (ymin > halfWin) binmin = ymin - halfWin;
|
Chris@275
|
256 else binmin = 0;
|
Chris@275
|
257
|
Chris@275
|
258 size_t binmax;
|
Chris@275
|
259 if (ymax + halfWin < values.size()) binmax = ymax + halfWin;
|
Chris@275
|
260 else binmax = values.size()-1;
|
Chris@275
|
261
|
Chris@275
|
262 for (size_t bin = binmin; bin <= binmax; ++bin) {
|
Chris@275
|
263
|
Chris@275
|
264 float value = values[bin];
|
Chris@275
|
265
|
Chris@275
|
266 window.push_back(value);
|
Chris@275
|
267
|
Chris@280
|
268 // so-called median will actually be the dist*100'th percentile
|
Chris@280
|
269 medianWinSize = getPeakPickWindowSize(type, sampleRate, bin, dist);
|
Chris@275
|
270 halfWin = medianWinSize/2;
|
Chris@275
|
271
|
Chris@275
|
272 while (window.size() > medianWinSize) window.pop_front();
|
Chris@275
|
273
|
Chris@275
|
274 if (type == MajorPitchAdaptivePeaks) {
|
Chris@275
|
275 if (ymax + halfWin < values.size()) binmax = ymax + halfWin;
|
Chris@275
|
276 else binmax = values.size()-1;
|
Chris@275
|
277 }
|
Chris@275
|
278
|
Chris@275
|
279 std::deque<float> sorted(window);
|
Chris@275
|
280 std::sort(sorted.begin(), sorted.end());
|
Chris@280
|
281 float median = sorted[int(sorted.size() * dist)];
|
Chris@275
|
282
|
Chris@275
|
283 if (value > median) {
|
Chris@275
|
284 inrange.push_back(bin);
|
Chris@275
|
285 }
|
Chris@275
|
286
|
Chris@275
|
287 if (value <= median || bin+1 == values.size()) {
|
Chris@275
|
288 size_t peakbin = 0;
|
Chris@275
|
289 float peakval = 0.f;
|
Chris@275
|
290 if (!inrange.empty()) {
|
Chris@275
|
291 for (size_t i = 0; i < inrange.size(); ++i) {
|
Chris@275
|
292 if (i == 0 || values[inrange[i]] > peakval) {
|
Chris@275
|
293 peakval = values[inrange[i]];
|
Chris@275
|
294 peakbin = inrange[i];
|
Chris@275
|
295 }
|
Chris@275
|
296 }
|
Chris@275
|
297 inrange.clear();
|
Chris@275
|
298 if (peakbin >= ymin && peakbin <= ymax) {
|
Chris@275
|
299 peaks.insert(peakbin);
|
Chris@275
|
300 }
|
Chris@275
|
301 }
|
Chris@275
|
302 }
|
Chris@275
|
303 }
|
Chris@275
|
304
|
Chris@275
|
305 return peaks;
|
Chris@275
|
306 }
|
Chris@275
|
307
|
Chris@275
|
308 size_t
|
Chris@280
|
309 FFTModel::getPeakPickWindowSize(PeakPickType type, size_t sampleRate,
|
Chris@280
|
310 size_t bin, float &percentile) const
|
Chris@275
|
311 {
|
Chris@280
|
312 percentile = 0.5;
|
Chris@275
|
313 if (type == MajorPeaks) return 10;
|
Chris@275
|
314 if (bin == 0) return 3;
|
Chris@280
|
315
|
Chris@275
|
316 size_t fftSize = m_server->getFFTSize() >> m_yshift;
|
Chris@275
|
317 float binfreq = (sampleRate * bin) / fftSize;
|
Chris@275
|
318 float hifreq = Pitch::getFrequencyForPitch(73, 0, binfreq);
|
Chris@280
|
319
|
Chris@275
|
320 int hibin = lrintf((hifreq * fftSize) / sampleRate);
|
Chris@275
|
321 int medianWinSize = hibin - bin;
|
Chris@275
|
322 if (medianWinSize < 3) medianWinSize = 3;
|
Chris@280
|
323
|
Chris@280
|
324 percentile = 0.5 + (binfreq / sampleRate);
|
Chris@280
|
325
|
Chris@275
|
326 return medianWinSize;
|
Chris@275
|
327 }
|
Chris@275
|
328
|
Chris@275
|
329 FFTModel::PeakSet
|
Chris@275
|
330 FFTModel::getPeakFrequencies(PeakPickType type, size_t x,
|
Chris@275
|
331 size_t ymin, size_t ymax)
|
Chris@275
|
332 {
|
Chris@275
|
333 PeakSet peaks;
|
Chris@275
|
334 if (!isOK()) return peaks;
|
Chris@275
|
335 PeakLocationSet locations = getPeaks(type, x, ymin, ymax);
|
Chris@275
|
336
|
Chris@275
|
337 size_t sampleRate = getSampleRate();
|
Chris@275
|
338 size_t fftSize = m_server->getFFTSize() >> m_yshift;
|
Chris@275
|
339 size_t incr = getResolution();
|
Chris@275
|
340
|
Chris@275
|
341 // This duplicates some of the work of estimateStableFrequency to
|
Chris@275
|
342 // allow us to retrieve the phases in two separate vertical
|
Chris@275
|
343 // columns, instead of jumping back and forth between columns x and
|
Chris@275
|
344 // x+1, which may be significantly slower if re-seeking is needed
|
Chris@275
|
345
|
Chris@275
|
346 std::vector<float> phases;
|
Chris@275
|
347 for (PeakLocationSet::iterator i = locations.begin();
|
Chris@275
|
348 i != locations.end(); ++i) {
|
Chris@275
|
349 phases.push_back(getPhaseAt(x, *i));
|
Chris@275
|
350 }
|
Chris@275
|
351
|
Chris@275
|
352 size_t phaseIndex = 0;
|
Chris@275
|
353 for (PeakLocationSet::iterator i = locations.begin();
|
Chris@275
|
354 i != locations.end(); ++i) {
|
Chris@275
|
355 float oldPhase = phases[phaseIndex];
|
Chris@275
|
356 float newPhase = getPhaseAt(x+1, *i);
|
Chris@275
|
357 float expectedPhase = oldPhase + (2.0 * M_PI * *i * incr) / fftSize;
|
Chris@275
|
358 float phaseError = princargf(newPhase - expectedPhase);
|
Chris@275
|
359 float frequency =
|
Chris@275
|
360 (sampleRate * (expectedPhase + phaseError - oldPhase))
|
Chris@275
|
361 / (2 * M_PI * incr);
|
Chris@275
|
362 // bool stable = (fabsf(phaseError) < (1.1f * (incr * M_PI) / fftSize));
|
Chris@275
|
363 // if (stable)
|
Chris@275
|
364 peaks[*i] = frequency;
|
Chris@275
|
365 ++phaseIndex;
|
Chris@275
|
366 }
|
Chris@275
|
367
|
Chris@275
|
368 return peaks;
|
Chris@275
|
369 }
|
Chris@275
|
370
|
Chris@152
|
371 Model *
|
Chris@152
|
372 FFTModel::clone() const
|
Chris@152
|
373 {
|
Chris@152
|
374 return new FFTModel(*this);
|
Chris@152
|
375 }
|
Chris@152
|
376
|
Chris@152
|
377 FFTModel::FFTModel(const FFTModel &model) :
|
Chris@152
|
378 DenseThreeDimensionalModel(),
|
Chris@152
|
379 m_server(model.m_server),
|
Chris@152
|
380 m_xshift(model.m_xshift),
|
Chris@152
|
381 m_yshift(model.m_yshift)
|
Chris@152
|
382 {
|
Chris@152
|
383 FFTDataServer::claimInstance(m_server);
|
Chris@152
|
384 }
|
Chris@152
|
385
|