Mercurial > hg > svcore
comparison transform/BeatDetectTransform.cpp @ 0:da6937383da8
initial import
author | Chris Cannam |
---|---|
date | Tue, 10 Jan 2006 16:33:16 +0000 |
parents | |
children | d86891498eef |
comparison
equal
deleted
inserted
replaced
-1:000000000000 | 0:da6937383da8 |
---|---|
1 /* -*- c-basic-offset: 4 -*- vi:set ts=8 sts=4 sw=4: */ | |
2 | |
3 /* | |
4 A waveform viewer and audio annotation editor. | |
5 Chris Cannam, Queen Mary University of London, 2005 | |
6 | |
7 This is experimental software. Not for distribution. | |
8 */ | |
9 | |
10 #include "BeatDetectTransform.h" | |
11 | |
12 #include "model/DenseTimeValueModel.h" | |
13 #include "model/SparseOneDimensionalModel.h" | |
14 | |
15 #include <iostream> | |
16 #include "dsp/onsets/DetectionFunction.h" | |
17 #include "dsp/tempotracking/TempoTrack.h" | |
18 | |
19 | |
20 BeatDetectTransform::BeatDetectTransform(Model *inputModel) : | |
21 Transform(inputModel) | |
22 { | |
23 // Step resolution for the detection function in seconds | |
24 double stepSecs = 0.01161; | |
25 | |
26 // Step resolution for the detection function in samples | |
27 size_t stepSize = (size_t)floor((double)inputModel->getSampleRate() * | |
28 stepSecs); | |
29 | |
30 | |
31 // m_w->m_bdf->setResolution(stepSize); | |
32 // output->setResolution(stepSize); | |
33 | |
34 std::cerr << "BeatDetectTransform::BeatDetectTransform: input sample rate " << inputModel->getSampleRate() << ", stepSecs " << stepSecs << ", stepSize " << stepSize << ", unrounded stepSize " << double(inputModel->getSampleRate()) * stepSecs << ", output sample rate " << inputModel->getSampleRate() / stepSize << ", unrounded output sample rate " << double(inputModel->getSampleRate()) / double(stepSize) << std::endl; | |
35 | |
36 m_output = new SparseOneDimensionalModel(inputModel->getSampleRate(), 1); | |
37 } | |
38 | |
39 BeatDetectTransform::~BeatDetectTransform() | |
40 { | |
41 // parent does it all | |
42 } | |
43 | |
44 TransformName | |
45 BeatDetectTransform::getName() | |
46 { | |
47 return tr("Beats"); | |
48 } | |
49 | |
50 void | |
51 BeatDetectTransform::run() | |
52 { | |
53 SparseOneDimensionalModel *output = getOutput(); | |
54 DenseTimeValueModel *input = getInput(); | |
55 if (!input) return; | |
56 | |
57 DFConfig config; | |
58 | |
59 config.DFType = DF_COMPLEXSD; | |
60 | |
61 // Step resolution for the detection function in seconds | |
62 config.stepSecs = 0.01161; | |
63 | |
64 // Step resolution for the detection function in samples | |
65 config.stepSize = (unsigned int)floor((double)input->getSampleRate() * | |
66 config.stepSecs ); | |
67 | |
68 config.frameLength = 2 * config.stepSize; | |
69 | |
70 unsigned int stepSize = config.stepSize; | |
71 unsigned int frameLength = config.frameLength; | |
72 | |
73 // m_w->m_bdf->setResolution(stepSize); | |
74 output->setResolution(stepSize); | |
75 | |
76 //Tempo Tracking Configuration Parameters | |
77 TTParams ttparams; | |
78 | |
79 // Low Pass filter coefficients for detection function smoothing | |
80 double* aCoeffs = new double[3]; | |
81 double* bCoeffs = new double[3]; | |
82 | |
83 aCoeffs[ 0 ] = 1; | |
84 aCoeffs[ 1 ] = -0.5949; | |
85 aCoeffs[ 2 ] = 0.2348; | |
86 bCoeffs[ 0 ] = 0.1600; | |
87 bCoeffs[ 1 ] = 0.3200; | |
88 bCoeffs[ 2 ] = 0.1600; | |
89 | |
90 ttparams.winLength = 512; | |
91 ttparams.lagLength = 128; | |
92 ttparams.LPOrd = 2; | |
93 ttparams.LPACoeffs = aCoeffs; | |
94 ttparams.LPBCoeffs = bCoeffs; | |
95 ttparams.alpha = 9; | |
96 ttparams.WinT.post = 8; | |
97 ttparams.WinT.pre = 7; | |
98 | |
99 //////////////////////////////////////////////////////////// | |
100 // DetectionFunction | |
101 //////////////////////////////////////////////////////////// | |
102 // Instantiate and configure detection function object | |
103 | |
104 DetectionFunction df(config); | |
105 | |
106 size_t origin = input->getStartFrame(); | |
107 size_t frameCount = input->getEndFrame() - origin; | |
108 size_t blocks = (frameCount / stepSize); | |
109 if (blocks * stepSize < frameCount) ++blocks; | |
110 | |
111 double *buffer = new double[frameLength]; | |
112 | |
113 // DF output with causal extension | |
114 unsigned int clen = blocks + ttparams.winLength; | |
115 double *dfOutput = new double[clen]; | |
116 | |
117 std::cerr << "Detecting beats at step size " << stepSize << "..." << std::endl; | |
118 | |
119 for (size_t i = 0; i < clen; ++i) { | |
120 | |
121 // std::cerr << "block " << i << "/" << clen << std::endl; | |
122 // std::cerr << "."; | |
123 | |
124 if (i < blocks) { | |
125 size_t got = input->getValues(-1, //!!! needs to come from parent layer -- which is not supposed to be in scope at this point | |
126 origin + i * stepSize, | |
127 origin + i * stepSize + frameLength, | |
128 buffer); | |
129 while (got < frameLength) buffer[got++] = 0.0; | |
130 dfOutput[i] = df.process(buffer); | |
131 } else { | |
132 dfOutput[i] = 0.0; | |
133 } | |
134 | |
135 // m_w->m_bdf->addPoint(SparseTimeValueModel::Point | |
136 // (i * stepSize, dfOutput[i], | |
137 // QString("%1").arg(dfOutput[i]))); | |
138 // m_w->m_bdf->setCompletion(i * 99 / clen); | |
139 output->setCompletion(i * 99 / clen); | |
140 | |
141 if (m_deleting) { | |
142 delete [] buffer; | |
143 delete [] dfOutput; | |
144 delete [] aCoeffs; | |
145 delete [] bCoeffs; | |
146 return; | |
147 } | |
148 } | |
149 | |
150 // m_w->m_bdf->setCompletion(100); | |
151 | |
152 // Tempo Track Object instantiation and configuration | |
153 TempoTrack tempoTracker(ttparams); | |
154 | |
155 // Vector of detected onsets | |
156 vector<int> beats; | |
157 | |
158 std::cerr << "Running tempo tracker..." << std::endl; | |
159 | |
160 beats = tempoTracker.process(dfOutput, blocks); | |
161 | |
162 delete [] buffer; | |
163 delete [] dfOutput; | |
164 delete [] aCoeffs; | |
165 delete [] bCoeffs; | |
166 | |
167 for (size_t i = 0; i < beats.size(); ++i) { | |
168 // std::cerr << "Beat value " << beats[i] << ", multiplying out to " << beats[i] * stepSize << std::endl; | |
169 float bpm = 0.0; | |
170 int fdiff = 0; | |
171 if (i < beats.size() - 1) { | |
172 fdiff = (beats[i+1] - beats[i]) * stepSize; | |
173 // one beat is fdiff frames, so there are samplerate/fdiff bps, | |
174 // so 60*samplerate/fdiff bpm | |
175 if (fdiff > 0) { | |
176 bpm = (60.0 * input->getSampleRate()) / fdiff; | |
177 } | |
178 } | |
179 output->addPoint(SparseOneDimensionalModel::Point | |
180 (origin + beats[i] * stepSize, QString("%1").arg(bpm))); | |
181 if (m_deleting) return; | |
182 } | |
183 | |
184 output->setCompletion(100); | |
185 } | |
186 | |
187 DenseTimeValueModel * | |
188 BeatDetectTransform::getInput() | |
189 { | |
190 DenseTimeValueModel *dtvm = | |
191 dynamic_cast<DenseTimeValueModel *>(getInputModel()); | |
192 if (!dtvm) { | |
193 std::cerr << "BeatDetectTransform::getInput: WARNING: Input model is not conformable to DenseTimeValueModel" << std::endl; | |
194 } | |
195 return dtvm; | |
196 } | |
197 | |
198 SparseOneDimensionalModel * | |
199 BeatDetectTransform::getOutput() | |
200 { | |
201 return static_cast<SparseOneDimensionalModel *>(getOutputModel()); | |
202 } | |
203 |