1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
|
{\rtf1\ansi\ansicpg1252\cocoartf1038\cocoasubrtf350
{\fonttbl\f0\fswiss\fcharset0 Helvetica;\f1\fnil\fcharset0 Monaco;\f2\froman\fcharset0 Times-Roman;
}
{\colortbl;\red255\green255\blue255;\red0\green0\blue191;\red0\green0\blue0;\red191\green0\blue0;
\red0\green0\blue191;\red96\green96\blue96;\red96\green96\blue96;\red0\green115\blue0;}
\deftab560
\pard\pardeftab560\ql\qnatural
\f0\b\fs36 \cf2 MFCC\cf3 mel frequency cepstral coefficients
\b0 \
\pard\pardeftab560\ql\qnatural
\fs24 \cf3 \
\b #coeff1, coeff2, ... = \cf2 MFCC\cf3 .kr(chain, numcoeff=13)
\b0 \
\
\b chain [fft] -
\b0 \cf2 Audio\cf3 input to track, which has been pre-analysed by the \cf2 FFT\cf3 \cf2 UGen\cf3 ; see examples below for the expected \cf2 FFT\cf3 size\
\b numcoeff [s] -
\b0 \cf2 Number\cf3 of coefficients, defaults to 13, maximum of 42; more efficient to use less of course! \
\
\cf2 Generates\cf3 a set of \cf2 MFCCs\cf3 ; these are obtained from a band-based frequency representation (using the \cf2 Mel\cf3 scale by default), and then a discrete cosine transform (\cf2 DCT\cf3 ). \cf2 The\cf3 \cf2 DCT\cf3 is an efficient approximation for principal components analysis, so that it allows a compression, or reduction of dimensionality, of the data, in \cf2 this\cf3 case reducing 42 band readings to a smaller set of \cf2 MFCCs\cf3 . \cf2 A\cf3 small number of features (the coefficients) end up describing the spectrum. \cf2 The\cf3 \cf2 MFCCs\cf3 are commonly used as timbral descriptors. \
\
\cf2 Output\cf3 values are somewhat normalised for the range 0.0 to 1.0, but there are no guarantees on exact conformance to \cf2 this\cf3 . \cf2 Commonly\cf3 , the first coefficient will be the highest value. \
\
\pard\pardeftab560\ql\qnatural
\f1\fs18 \cf4 //Technical note: The 0th coefficient is not generated as it consists of multiplying all bands by 1 and summing\cf3 \
\
\
\cf4 //assumes hop of half fftsize, fine\cf3 \
b = \cf2 Buffer\cf3 .alloc(s,1024,1); \cf4 //for sampling rates 44100 and 48000\cf3 \
\cf4 //b = Buffer.alloc(s,2048,1); //for sampling rates 88200 and 96000\cf3 \
\cf0 d=\cf5 Buffer\cf0 .read(s,\cf5 Document\cf0 .current.dir ++ \cf6 "/pinknoise2.wav"\cf0 );\
d=\cf5 Buffer\cf0 .read(s,\cf5 Document\cf0 .current.dir ++ \cf6 "/whitenoise2.wav"\cf0 );\
\cf3 d=\cf2 Buffer\cf3 .read(s,\cf2 Document\cf3 .current.dir ++ \cf7 "/sinetest_stepping2.wav"\cf3 );\
d.play\
d\
(\
x= \{\
\cf2 var\cf3 in, fft, array;\
\
in= \cf2 PlayBuf\cf3 .ar(1,d,\cf2 BufRateScale\cf3 .kr(d),1,0,1);\
\
\cf4 //in = SoundIn.ar(0); \cf3 \
\
fft = \cf2 FFT\cf3 (b, in, 512, 1);\
\
array=\cf2 MFCC\cf3 .kr(fft); \
\
array.size.postln; \
\
\cf2 Out\cf3 .kr(0,array); \
\
\cf2 Out\cf3 .ar(0,\cf2 Pan2\cf3 .ar(in)); \
\}.play\
)\
\
\
c= \cf2 Bus\cf3 .new(\cf8 'control'\cf3 , 0, 13); \
\
\cf4 //poll coefficients\cf3 \
c.getn(13,\{\cf2 arg\cf3 val; \{val.plot;\}.defer\}); \
\
\
\
\cf4 //Continuous graphical display of MFCC values; free routine before closing window\cf3 \
\
(\
\cf2 var\cf3 ms; \
\
w=\cf2 Window\cf3 .new(\cf7 "Thirteen MFCC coefficients"\cf3 , \cf2 Rect\cf3 (200,400,700,300));\
\
ms= \cf2 MultiSliderView\cf3 .new(w, \cf2 Rect\cf3 (10,10,660,280));\
\
ms.value_(\cf2 Array\cf3 .fill(13,0.0));\
ms.valueThumbSize_(40.0);\
ms.indexThumbSize_(40.0);\
ms.gap_(0);\
\
w.front;\
\
r= \{\
\
\cf2 inf\cf3 .do\{\
\
c.getn(13,\{\cf2 arg\cf3 val; val.postln; \{ms.value_(val*0.9)\}.defer\}); \
\
0.02.wait; \cf4 //25 frames per second\cf3 \
\};\
\
\}.fork;\
\
)\
\
\
\cf4 //tidy up\cf3 \
(\
r.stop;\
b.free;\
c.free;\
x.free;\
w.close;\
)\
\pard\pardeftab560\ql\qnatural
\f2\fs32 \cf3 \
\
\
\
}
|
