Chris@9
|
1 /* -*- c-basic-offset: 4 indent-tabs-mode: nil -*- vi:set ts=8 sts=4 sw=4: */
|
Chris@9
|
2
|
Chris@9
|
3 /*
|
Chris@9
|
4 pYIN - A fundamental frequency estimator for monophonic audio
|
Chris@9
|
5 Centre for Digital Music, Queen Mary, University of London.
|
Chris@9
|
6
|
Chris@9
|
7 This program is free software; you can redistribute it and/or
|
Chris@9
|
8 modify it under the terms of the GNU General Public License as
|
Chris@9
|
9 published by the Free Software Foundation; either version 2 of the
|
Chris@9
|
10 License, or (at your option) any later version. See the file
|
Chris@9
|
11 COPYING included with this distribution for more information.
|
Chris@9
|
12 */
|
Chris@9
|
13
|
matthiasm@0
|
14 #include "MonoNote.h"
|
matthiasm@0
|
15 #include <vector>
|
matthiasm@0
|
16
|
matthiasm@0
|
17 #include <cstdio>
|
matthiasm@0
|
18 #include <cmath>
|
matthiasm@0
|
19 #include <complex>
|
matthiasm@0
|
20
|
matthiasm@0
|
21 using std::vector;
|
matthiasm@0
|
22 using std::pair;
|
matthiasm@0
|
23
|
matthiasm@0
|
24 MonoNote::MonoNote() :
|
mail@132
|
25 hmm(0)
|
matthiasm@0
|
26 {
|
matthiasm@0
|
27 }
|
matthiasm@0
|
28
|
matthiasm@0
|
29 MonoNote::~MonoNote()
|
matthiasm@0
|
30 {
|
matthiasm@0
|
31 }
|
matthiasm@0
|
32
|
matthiasm@0
|
33 const vector<MonoNote::FrameOutput>
|
matthiasm@0
|
34 MonoNote::process(const vector<vector<pair<double, double> > > pitchProb)
|
matthiasm@0
|
35 {
|
Chris@146
|
36 // Previously, this built up a single matrix of probabilities, by
|
Chris@146
|
37 // calling calculateObsProb to get a column for each frame in
|
Chris@146
|
38 // pitchProb.
|
Chris@146
|
39 //
|
Chris@146
|
40 // The number of distinct states depends on MonoNoteParameters,
|
Chris@146
|
41 // but the defaults have 3 states per pitch, 3 pitches per MIDI
|
Chris@146
|
42 // note, and 69 MIDI notes, giving 681 states per frame. With a
|
Chris@146
|
43 // frame step size of 256 at 44100Hz sample rate, a 3-minute song
|
Chris@146
|
44 // has about 30K frames leading to a 20 million element
|
Chris@146
|
45 // probability matrix.
|
Chris@146
|
46 //
|
Chris@146
|
47 // Since the matrix is very sparse, we can avoid some of this by
|
Chris@146
|
48 // feeding the (sparse implementation of) HMM one column at a
|
Chris@146
|
49 // time.
|
Chris@146
|
50
|
Chris@146
|
51 vector<int> path;
|
Chris@146
|
52
|
Chris@146
|
53 if (!pitchProb.empty()) {
|
Chris@146
|
54
|
Chris@146
|
55 hmm.initialise(hmm.calculateObsProb(pitchProb[0]));
|
Chris@146
|
56
|
Chris@146
|
57 for (size_t iFrame = 1; iFrame < pitchProb.size(); ++iFrame)
|
Chris@146
|
58 {
|
Chris@146
|
59 hmm.process(hmm.calculateObsProb(pitchProb[iFrame]));
|
Chris@146
|
60 }
|
Chris@146
|
61
|
Chris@146
|
62 path = hmm.track();
|
matthiasm@0
|
63 }
|
matthiasm@0
|
64
|
matthiasm@0
|
65 vector<MonoNote::FrameOutput> out;
|
Chris@146
|
66
|
matthiasm@0
|
67 for (size_t iFrame = 0; iFrame < path.size(); ++iFrame)
|
matthiasm@0
|
68 {
|
matthiasm@0
|
69 double currPitch = -1;
|
matthiasm@0
|
70 int stateKind = 0;
|
matthiasm@0
|
71
|
matthiasm@0
|
72 currPitch = hmm.par.minPitch + (path[iFrame]/hmm.par.nSPP) * 1.0/hmm.par.nPPS;
|
matthiasm@0
|
73 stateKind = (path[iFrame]) % hmm.par.nSPP + 1;
|
matthiasm@0
|
74
|
matthiasm@0
|
75 out.push_back(FrameOutput(iFrame, currPitch, stateKind));
|
matthiasm@0
|
76 }
|
Chris@146
|
77
|
matthiasm@0
|
78 return(out);
|
Chris@9
|
79 }
|