Chris@0
|
1 /* -*- c-basic-offset: 4 indent-tabs-mode: nil -*- vi:set ts=8 sts=4 sw=4: */
|
Chris@0
|
2
|
Chris@0
|
3 /*
|
Chris@0
|
4 Sonic Visualiser
|
Chris@0
|
5 An audio file viewer and annotation editor.
|
Chris@0
|
6 Centre for Digital Music, Queen Mary, University of London.
|
Chris@0
|
7 This file copyright 2006 Chris Cannam.
|
Chris@0
|
8
|
Chris@0
|
9 This program is free software; you can redistribute it and/or
|
Chris@0
|
10 modify it under the terms of the GNU General Public License as
|
Chris@0
|
11 published by the Free Software Foundation; either version 2 of the
|
Chris@0
|
12 License, or (at your option) any later version. See the file
|
Chris@0
|
13 COPYING included with this distribution for more information.
|
Chris@0
|
14 */
|
Chris@0
|
15
|
Chris@14
|
16 #ifndef _PHASE_VOCODER_TIME_STRETCHER_H_
|
Chris@14
|
17 #define _PHASE_VOCODER_TIME_STRETCHER_H_
|
Chris@0
|
18
|
Chris@0
|
19 #include "base/Window.h"
|
Chris@0
|
20 #include "base/RingBuffer.h"
|
Chris@0
|
21
|
Chris@0
|
22 #include <fftw3.h>
|
Chris@0
|
23
|
Chris@0
|
24 /**
|
Chris@14
|
25 * A time stretcher that alters the performance speed of audio,
|
Chris@14
|
26 * preserving pitch. This uses the simple phase vocoder technique
|
Chris@14
|
27 * from DAFX pp275-276, adding a block-based stream oriented API.
|
Chris@0
|
28 *
|
Chris@0
|
29 * Causes significant transient smearing, but sounds good for steady
|
Chris@0
|
30 * notes and is generally predictable.
|
Chris@0
|
31 */
|
Chris@0
|
32
|
Chris@14
|
33 class PhaseVocoderTimeStretcher
|
Chris@0
|
34 {
|
Chris@0
|
35 public:
|
Chris@16
|
36 PhaseVocoderTimeStretcher(size_t channels,
|
Chris@16
|
37 float ratio,
|
Chris@16
|
38 bool sharpen,
|
Chris@16
|
39 size_t maxProcessInputBlockSize);
|
Chris@14
|
40 virtual ~PhaseVocoderTimeStretcher();
|
Chris@0
|
41
|
Chris@12
|
42 /**
|
Chris@12
|
43 * Process a block. The input array contains the given number of
|
Chris@16
|
44 * samples (on each channel); the output must have space for
|
Chris@16
|
45 * lrintf(samples * m_ratio).
|
Chris@16
|
46 *
|
Chris@16
|
47 * This should work correctly for some ratios, e.g. small powers
|
Chris@16
|
48 * of two. For other ratios it may drop samples -- use putInput
|
Chris@16
|
49 * in a loop followed by getOutput (when getAvailableOutputSamples
|
Chris@16
|
50 * reports enough) instead.
|
Chris@16
|
51 *
|
Chris@16
|
52 * Do not mix process calls with putInput/getOutput calls.
|
Chris@12
|
53 */
|
Chris@16
|
54 void process(float **input, float **output, size_t samples);
|
Chris@16
|
55
|
Chris@16
|
56 /**
|
Chris@16
|
57 * Return the number of samples that would need to be added via
|
Chris@16
|
58 * putInput in order to provoke the time stretcher into doing some
|
Chris@16
|
59 * time stretching and making more output samples available.
|
Chris@16
|
60 */
|
Chris@16
|
61 size_t getRequiredInputSamples() const;
|
Chris@16
|
62
|
Chris@16
|
63 /**
|
Chris@16
|
64 * Put (and possibly process) a given number of input samples.
|
Chris@16
|
65 * Number must not exceed the maxProcessInputBlockSize passed to
|
Chris@16
|
66 * constructor.
|
Chris@16
|
67 */
|
Chris@16
|
68 void putInput(float **input, size_t samples);
|
Chris@16
|
69
|
Chris@16
|
70 size_t getAvailableOutputSamples() const;
|
Chris@16
|
71
|
Chris@16
|
72 void getOutput(float **output, size_t samples);
|
Chris@16
|
73
|
Chris@16
|
74 //!!! and reset?
|
Chris@0
|
75
|
Chris@0
|
76 /**
|
Chris@15
|
77 * Get the hop size for input.
|
Chris@0
|
78 */
|
Chris@0
|
79 size_t getInputIncrement() const { return m_n1; }
|
Chris@0
|
80
|
Chris@0
|
81 /**
|
Chris@15
|
82 * Get the hop size for output.
|
Chris@15
|
83 */
|
Chris@16
|
84 size_t getOutputIncrement() const { return m_n2; }
|
Chris@15
|
85
|
Chris@15
|
86 /**
|
Chris@15
|
87 * Get the window size for FFT processing.
|
Chris@0
|
88 */
|
Chris@0
|
89 size_t getWindowSize() const { return m_wlen; }
|
Chris@0
|
90
|
Chris@0
|
91 /**
|
Chris@15
|
92 * Get the window type.
|
Chris@0
|
93 */
|
Chris@20
|
94 // WindowType getWindowType() const { return m_window->getType(); }
|
Chris@0
|
95
|
Chris@15
|
96 /**
|
Chris@16
|
97 * Get the stretch ratio.
|
Chris@15
|
98 */
|
Chris@16
|
99 float getRatio() const { return float(m_n2) / float(m_n1); }
|
Chris@16
|
100
|
Chris@16
|
101 /**
|
Chris@16
|
102 * Return whether this time stretcher will attempt to sharpen transients.
|
Chris@16
|
103 */
|
Chris@16
|
104 bool getSharpening() const { return m_sharpen; }
|
Chris@15
|
105
|
Chris@15
|
106 /**
|
Chris@15
|
107 * Get the latency added by the time stretcher, in sample frames.
|
Chris@15
|
108 */
|
Chris@0
|
109 size_t getProcessingLatency() const;
|
Chris@0
|
110
|
Chris@0
|
111 protected:
|
Chris@13
|
112 /**
|
Chris@13
|
113 * Process a single phase vocoder frame.
|
Chris@13
|
114 *
|
Chris@13
|
115 * Take m_wlen time-domain source samples from in, perform an FFT,
|
Chris@13
|
116 * phase shift, and IFFT, and add the results to out (presumably
|
Chris@13
|
117 * overlapping parts of existing data from prior frames).
|
Chris@13
|
118 *
|
Chris@13
|
119 * Also add to the modulation output the results of windowing a
|
Chris@13
|
120 * set of 1s with the resynthesis window -- this can then be used
|
Chris@13
|
121 * to ensure the output has the correct magnitude in cases where
|
Chris@13
|
122 * the window overlap varies or otherwise results in something
|
Chris@13
|
123 * other than a flat sum.
|
Chris@13
|
124 */
|
Chris@20
|
125
|
Chris@20
|
126
|
Chris@20
|
127 void analyseBlock(size_t channel, float *in); // into m_freq[channel]
|
Chris@20
|
128
|
Chris@20
|
129 bool isTransient(); // operates on m_freq[0..m_channels-1]
|
Chris@20
|
130
|
Chris@20
|
131 void synthesiseBlock(size_t channel, float *out, float *modulation,
|
Chris@20
|
132 size_t lastStep);
|
Chris@0
|
133
|
Chris@16
|
134 size_t m_channels;
|
Chris@12
|
135 float m_ratio;
|
Chris@16
|
136 bool m_sharpen;
|
Chris@0
|
137 size_t m_n1;
|
Chris@0
|
138 size_t m_n2;
|
Chris@0
|
139 size_t m_wlen;
|
Chris@20
|
140 Window<float> *m_analysisWindow;
|
Chris@20
|
141 Window<float> *m_synthesisWindow;
|
Chris@0
|
142
|
Chris@16
|
143 float **m_prevPhase;
|
Chris@16
|
144 float **m_prevAdjustedPhase;
|
Chris@16
|
145
|
Chris@20
|
146 float *m_prevTransientMag;
|
Chris@20
|
147 int m_prevTransientCount;
|
Chris@20
|
148 bool m_prevTransient;
|
Chris@20
|
149
|
Chris@20
|
150 float *m_tempbuf;
|
Chris@20
|
151 float **m_time;
|
Chris@20
|
152 fftwf_complex **m_freq;
|
Chris@20
|
153 fftwf_plan *m_plan;
|
Chris@20
|
154 fftwf_plan *m_iplan;
|
Chris@0
|
155
|
Chris@16
|
156 RingBuffer<float> **m_inbuf;
|
Chris@16
|
157 RingBuffer<float> **m_outbuf;
|
Chris@16
|
158 float **m_mashbuf;
|
Chris@13
|
159 float *m_modulationbuf;
|
Chris@0
|
160 };
|
Chris@0
|
161
|
Chris@0
|
162 #endif
|