c@225
|
1 /* -*- c-basic-offset: 4 indent-tabs-mode: nil -*- vi:set ts=8 sts=4 sw=4: */
|
c@225
|
2
|
c@225
|
3 /*
|
c@225
|
4 QM DSP Library
|
c@225
|
5
|
c@225
|
6 Centre for Digital Music, Queen Mary, University of London.
|
c@309
|
7 This file 2005-2006 Christian Landone.
|
c@309
|
8
|
c@309
|
9 This program is free software; you can redistribute it and/or
|
c@309
|
10 modify it under the terms of the GNU General Public License as
|
c@309
|
11 published by the Free Software Foundation; either version 2 of the
|
c@309
|
12 License, or (at your option) any later version. See the file
|
c@309
|
13 COPYING included with this distribution for more information.
|
c@225
|
14 */
|
c@225
|
15
|
c@225
|
16 #include "DetectionFunction.h"
|
c@272
|
17 #include <cstring>
|
c@225
|
18
|
c@225
|
19 //////////////////////////////////////////////////////////////////////
|
c@225
|
20 // Construction/Destruction
|
c@225
|
21 //////////////////////////////////////////////////////////////////////
|
c@225
|
22
|
c@225
|
23 DetectionFunction::DetectionFunction( DFConfig Config ) :
|
c@225
|
24 m_window(0)
|
c@225
|
25 {
|
c@227
|
26 m_magHistory = NULL;
|
c@227
|
27 m_phaseHistory = NULL;
|
c@227
|
28 m_phaseHistoryOld = NULL;
|
c@239
|
29 m_magPeaks = NULL;
|
c@225
|
30
|
c@225
|
31 initialise( Config );
|
c@225
|
32 }
|
c@225
|
33
|
c@225
|
34 DetectionFunction::~DetectionFunction()
|
c@225
|
35 {
|
c@225
|
36 deInitialise();
|
c@225
|
37 }
|
c@225
|
38
|
c@225
|
39
|
c@225
|
40 void DetectionFunction::initialise( DFConfig Config )
|
c@225
|
41 {
|
c@225
|
42 m_dataLength = Config.frameLength;
|
c@340
|
43 m_halfLength = m_dataLength/2 + 1;
|
c@239
|
44
|
c@225
|
45 m_DFType = Config.DFType;
|
c@238
|
46 m_stepSize = Config.stepSize;
|
c@225
|
47
|
c@239
|
48 m_whiten = Config.adaptiveWhitening;
|
c@239
|
49 m_whitenRelaxCoeff = Config.whiteningRelaxCoeff;
|
c@239
|
50 m_whitenFloor = Config.whiteningFloor;
|
c@239
|
51 if (m_whitenRelaxCoeff < 0) m_whitenRelaxCoeff = 0.9997;
|
c@239
|
52 if (m_whitenFloor < 0) m_whitenFloor = 0.01;
|
c@239
|
53
|
c@227
|
54 m_magHistory = new double[ m_halfLength ];
|
c@227
|
55 memset(m_magHistory,0, m_halfLength*sizeof(double));
|
c@225
|
56
|
c@227
|
57 m_phaseHistory = new double[ m_halfLength ];
|
c@227
|
58 memset(m_phaseHistory,0, m_halfLength*sizeof(double));
|
c@225
|
59
|
c@227
|
60 m_phaseHistoryOld = new double[ m_halfLength ];
|
c@227
|
61 memset(m_phaseHistoryOld,0, m_halfLength*sizeof(double));
|
c@225
|
62
|
c@239
|
63 m_magPeaks = new double[ m_halfLength ];
|
c@239
|
64 memset(m_magPeaks,0, m_halfLength*sizeof(double));
|
c@239
|
65
|
c@356
|
66 m_phaseVoc = new PhaseVocoder(m_dataLength, m_stepSize);
|
c@225
|
67
|
c@225
|
68 m_magnitude = new double[ m_halfLength ];
|
c@225
|
69 m_thetaAngle = new double[ m_halfLength ];
|
c@340
|
70 m_unwrapped = new double[ m_halfLength ];
|
c@225
|
71
|
c@225
|
72 m_window = new Window<double>(HanningWindow, m_dataLength);
|
c@344
|
73 m_windowed = new double[ m_dataLength ];
|
c@225
|
74 }
|
c@225
|
75
|
c@225
|
76 void DetectionFunction::deInitialise()
|
c@225
|
77 {
|
c@227
|
78 delete [] m_magHistory ;
|
c@227
|
79 delete [] m_phaseHistory ;
|
c@227
|
80 delete [] m_phaseHistoryOld ;
|
c@239
|
81 delete [] m_magPeaks ;
|
c@225
|
82
|
c@225
|
83 delete m_phaseVoc;
|
c@225
|
84
|
c@225
|
85 delete [] m_magnitude;
|
c@225
|
86 delete [] m_thetaAngle;
|
c@344
|
87 delete [] m_windowed;
|
c@344
|
88 delete [] m_unwrapped;
|
c@225
|
89
|
c@225
|
90 delete m_window;
|
c@225
|
91 }
|
c@225
|
92
|
c@344
|
93 double DetectionFunction::processTimeDomain(const double *samples)
|
c@225
|
94 {
|
c@344
|
95 m_window->cut(samples, m_windowed);
|
c@280
|
96
|
c@344
|
97 m_phaseVoc->processTimeDomain(m_windowed,
|
c@344
|
98 m_magnitude, m_thetaAngle, m_unwrapped);
|
c@225
|
99
|
c@239
|
100 if (m_whiten) whiten();
|
c@239
|
101
|
c@227
|
102 return runDF();
|
c@227
|
103 }
|
c@227
|
104
|
c@344
|
105 double DetectionFunction::processFrequencyDomain(const double *reals,
|
c@344
|
106 const double *imags)
|
c@227
|
107 {
|
c@344
|
108 m_phaseVoc->processFrequencyDomain(reals, imags,
|
c@344
|
109 m_magnitude, m_thetaAngle, m_unwrapped);
|
c@227
|
110
|
c@239
|
111 if (m_whiten) whiten();
|
c@239
|
112
|
c@227
|
113 return runDF();
|
c@227
|
114 }
|
c@227
|
115
|
c@239
|
116 void DetectionFunction::whiten()
|
c@239
|
117 {
|
c@239
|
118 for (unsigned int i = 0; i < m_halfLength; ++i) {
|
c@239
|
119 double m = m_magnitude[i];
|
c@239
|
120 if (m < m_magPeaks[i]) {
|
c@239
|
121 m = m + (m_magPeaks[i] - m) * m_whitenRelaxCoeff;
|
c@239
|
122 }
|
c@239
|
123 if (m < m_whitenFloor) m = m_whitenFloor;
|
c@239
|
124 m_magPeaks[i] = m;
|
c@239
|
125 m_magnitude[i] /= m;
|
c@239
|
126 }
|
c@239
|
127 }
|
c@239
|
128
|
c@227
|
129 double DetectionFunction::runDF()
|
c@227
|
130 {
|
c@227
|
131 double retVal = 0;
|
c@227
|
132
|
c@225
|
133 switch( m_DFType )
|
c@225
|
134 {
|
c@225
|
135 case DF_HFC:
|
c@225
|
136 retVal = HFC( m_halfLength, m_magnitude);
|
c@225
|
137 break;
|
c@225
|
138
|
c@238
|
139 case DF_SPECDIFF:
|
c@225
|
140 retVal = specDiff( m_halfLength, m_magnitude);
|
c@225
|
141 break;
|
c@225
|
142
|
c@225
|
143 case DF_PHASEDEV:
|
c@345
|
144 // Using the instantaneous phases here actually provides the
|
c@345
|
145 // same results (for these calculations) as if we had used
|
c@345
|
146 // unwrapped phases, but without the possible accumulation of
|
c@345
|
147 // phase error over time
|
c@239
|
148 retVal = phaseDev( m_halfLength, m_thetaAngle);
|
c@225
|
149 break;
|
c@225
|
150
|
c@225
|
151 case DF_COMPLEXSD:
|
c@225
|
152 retVal = complexSD( m_halfLength, m_magnitude, m_thetaAngle);
|
c@225
|
153 break;
|
c@237
|
154
|
c@237
|
155 case DF_BROADBAND:
|
c@239
|
156 retVal = broadband( m_halfLength, m_magnitude);
|
c@239
|
157 break;
|
c@225
|
158 }
|
c@225
|
159
|
c@225
|
160 return retVal;
|
c@225
|
161 }
|
c@225
|
162
|
c@225
|
163 double DetectionFunction::HFC(unsigned int length, double *src)
|
c@225
|
164 {
|
c@225
|
165 unsigned int i;
|
c@225
|
166 double val = 0;
|
c@225
|
167
|
c@225
|
168 for( i = 0; i < length; i++)
|
c@225
|
169 {
|
c@225
|
170 val += src[ i ] * ( i + 1);
|
c@225
|
171 }
|
c@225
|
172 return val;
|
c@225
|
173 }
|
c@225
|
174
|
c@225
|
175 double DetectionFunction::specDiff(unsigned int length, double *src)
|
c@225
|
176 {
|
c@225
|
177 unsigned int i;
|
c@225
|
178 double val = 0.0;
|
c@225
|
179 double temp = 0.0;
|
c@225
|
180 double diff = 0.0;
|
c@225
|
181
|
c@225
|
182 for( i = 0; i < length; i++)
|
c@225
|
183 {
|
c@227
|
184 temp = fabs( (src[ i ] * src[ i ]) - (m_magHistory[ i ] * m_magHistory[ i ]) );
|
c@225
|
185
|
c@225
|
186 diff= sqrt(temp);
|
c@225
|
187
|
c@238
|
188 // (See note in phaseDev below.)
|
c@238
|
189
|
c@238
|
190 val += diff;
|
c@225
|
191
|
c@227
|
192 m_magHistory[ i ] = src[ i ];
|
c@225
|
193 }
|
c@225
|
194
|
c@225
|
195 return val;
|
c@225
|
196 }
|
c@225
|
197
|
c@225
|
198
|
c@239
|
199 double DetectionFunction::phaseDev(unsigned int length, double *srcPhase)
|
c@225
|
200 {
|
c@225
|
201 unsigned int i;
|
c@225
|
202 double tmpPhase = 0;
|
c@225
|
203 double tmpVal = 0;
|
c@225
|
204 double val = 0;
|
c@225
|
205
|
c@225
|
206 double dev = 0;
|
c@225
|
207
|
c@225
|
208 for( i = 0; i < length; i++)
|
c@225
|
209 {
|
c@227
|
210 tmpPhase = (srcPhase[ i ]- 2*m_phaseHistory[ i ]+m_phaseHistoryOld[ i ]);
|
c@225
|
211 dev = MathUtilities::princarg( tmpPhase );
|
c@238
|
212
|
c@238
|
213 // A previous version of this code only counted the value here
|
c@238
|
214 // if the magnitude exceeded 0.1. My impression is that
|
c@238
|
215 // doesn't greatly improve the results for "loud" music (so
|
c@238
|
216 // long as the peak picker is reasonably sophisticated), but
|
c@238
|
217 // does significantly damage its ability to work with quieter
|
c@238
|
218 // music, so I'm removing it and counting the result always.
|
c@238
|
219 // Same goes for the spectral difference measure above.
|
c@225
|
220
|
c@238
|
221 tmpVal = fabs(dev);
|
c@238
|
222 val += tmpVal ;
|
c@225
|
223
|
c@227
|
224 m_phaseHistoryOld[ i ] = m_phaseHistory[ i ] ;
|
c@227
|
225 m_phaseHistory[ i ] = srcPhase[ i ];
|
c@225
|
226 }
|
c@225
|
227
|
c@225
|
228 return val;
|
c@225
|
229 }
|
c@225
|
230
|
c@225
|
231
|
c@225
|
232 double DetectionFunction::complexSD(unsigned int length, double *srcMagnitude, double *srcPhase)
|
c@225
|
233 {
|
c@225
|
234 unsigned int i;
|
c@225
|
235 double val = 0;
|
c@225
|
236 double tmpPhase = 0;
|
c@225
|
237 double tmpReal = 0;
|
c@225
|
238 double tmpImag = 0;
|
c@225
|
239
|
c@225
|
240 double dev = 0;
|
c@225
|
241 ComplexData meas = ComplexData( 0, 0 );
|
c@227
|
242 ComplexData j = ComplexData( 0, 1 );
|
c@225
|
243
|
c@225
|
244 for( i = 0; i < length; i++)
|
c@225
|
245 {
|
c@227
|
246 tmpPhase = (srcPhase[ i ]- 2*m_phaseHistory[ i ]+m_phaseHistoryOld[ i ]);
|
c@225
|
247 dev= MathUtilities::princarg( tmpPhase );
|
c@225
|
248
|
c@227
|
249 meas = m_magHistory[i] - ( srcMagnitude[ i ] * exp( j * dev) );
|
c@225
|
250
|
c@225
|
251 tmpReal = real( meas );
|
c@225
|
252 tmpImag = imag( meas );
|
c@225
|
253
|
c@225
|
254 val += sqrt( (tmpReal * tmpReal) + (tmpImag * tmpImag) );
|
c@225
|
255
|
c@227
|
256 m_phaseHistoryOld[ i ] = m_phaseHistory[ i ] ;
|
c@227
|
257 m_phaseHistory[ i ] = srcPhase[ i ];
|
c@227
|
258 m_magHistory[ i ] = srcMagnitude[ i ];
|
c@225
|
259 }
|
c@225
|
260
|
c@225
|
261 return val;
|
c@225
|
262 }
|
c@225
|
263
|
c@239
|
264 double DetectionFunction::broadband(unsigned int length, double *src)
|
c@237
|
265 {
|
c@237
|
266 double val = 0;
|
c@237
|
267 for (unsigned int i = 0; i < length; ++i) {
|
c@239
|
268 double sqrmag = src[i] * src[i];
|
c@237
|
269 if (m_magHistory[i] > 0.0) {
|
c@237
|
270 double diff = 10.0 * log10(sqrmag / m_magHistory[i]);
|
c@237
|
271 if (diff > m_dbRise) val = val + 1;
|
c@237
|
272 }
|
c@237
|
273 m_magHistory[i] = sqrmag;
|
c@237
|
274 }
|
c@237
|
275 return val;
|
c@237
|
276 }
|
c@237
|
277
|
c@225
|
278 double* DetectionFunction::getSpectrumMagnitude()
|
c@225
|
279 {
|
c@225
|
280 return m_magnitude;
|
c@225
|
281 }
|
c@225
|
282
|