Mercurial > hg > camir-aes2014

annotate toolboxes/MIRtoolbox1.3.2/MIRToolboxDemos/demo4segmentation.m @ 0:e9a9cd732c1e tip

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
first hg version after svn
author wolffd
date Tue, 10 Feb 2015 15:05:51 +0000
parents
children
rev   line source
wolffd@0 1 function demo4segmentation
wolffd@0 2 % To get familiar with some approaches of segmentation of audio files
wolffd@0 3 % using MIRtoolbox.
wolffd@0 4
wolffd@0 5 % 1. Load an audio file (for instance, guitar.wav).
wolffd@0 6 a = miraudio('guitar');
wolffd@0 7
wolffd@0 8 % 2. We will perform the segmentation strategy as proposed in (Foote &
wolffd@0 9 % Cooper, 2003). First, decompose the file into successive frames of 50 ms
wolffd@0 10 % without overlap.
wolffd@0 11 help mirframe
wolffd@0 12 fr = mirframe(a,0.05,1)
wolffd@0 13
wolffd@0 14 % 3. Compute the spectrum representation (FFT) of the frames.
wolffd@0 15 sp = mirspectrum(fr)
wolffd@0 16 clear fr
wolffd@0 17 % (Remove from the memory any data that will not be used any more.)
wolffd@0 18
wolffd@0 19 % 4. Compute the similarity matrix that shows the similarity between the
wolffd@0 20 % spectrum of different frames.
wolffd@0 21 help mirsimatrix
wolffd@0 22 sm = mirsimatrix(sp)
wolffd@0 23 clear sp
wolffd@0 24 % Look at the structures shown in the matrix and find the relation with the
wolffd@0 25 % structure heard when listening to the extract.
wolffd@0 26
wolffd@0 27 % 5. Estimate the novelty score related to the similarity matrix. It
wolffd@0 28 % consists in a convolution of the diagonal of the matrix with a
wolffd@0 29 % checker-board Gaussian kernel. Use the novelty function for that purpose.
wolffd@0 30 help mirnovelty
wolffd@0 31 nv = mirnovelty(sm)
wolffd@0 32
wolffd@0 33 % 6. Detect the peaks in the novelty score.
wolffd@0 34 help mirpeaks
wolffd@0 35 p1 = mirpeaks(nv)
wolffd@0 36
wolffd@0 37 % You can change the threshold value of the peak picker function in order to
wolffd@0 38 % get better results.
wolffd@0 39 p2 = mirpeaks(nv,'Contrast',0.01)
wolffd@0 40
wolffd@0 41 clear nv
wolffd@0 42
wolffd@0 43 % 7. Segment the original audio file using the peaks as position for
wolffd@0 44 % segmentation.
wolffd@0 45 help mirsegment
wolffd@0 46 s1 = mirsegment(a,p1)
wolffd@0 47 clear p1
wolffd@0 48
wolffd@0 49 % 8. Listen to the results.
wolffd@0 50 mirplay(s1)
wolffd@0 51
wolffd@0 52 %s2 = mirsegment(a,p2)
wolffd@0 53 %clear p2
wolffd@0 54 %mirplay(s2)
wolffd@0 55
wolffd@0 56 % 9. Compute the similarity matrix of this obtained segmentation, in order
wolffd@0 57 % to view the relationships between the different segments and their
wolffd@0 58 % possible clustering into higher-level groups.
wolffd@0 59 mirsimatrix(s1,'Similarity')
wolffd@0 60 clear s1
wolffd@0 61 %mirsimatrix(s2)
wolffd@0 62 %clear s2
wolffd@0 63
wolffd@0 64 display('Strike any key to continue...');
wolffd@0 65 pause
wolffd@0 66 close all
wolffd@0 67
wolffd@0 68 % 10. Change the size of the kernel used in the novelty function, in order
wolffd@0 69 % to obtain segmentations of different levels of detail, from detailed
wolffd@0 70 % analysis of the local texture, to very simple segmentation of the whole
wolffd@0 71 % piece.
wolffd@0 72 n100 = mirnovelty(sm,'KernelSize',100)
wolffd@0 73 n50 = mirnovelty(sm,'KernelSize',50)
wolffd@0 74 n10 = mirnovelty(sm,'KernelSize',10)
wolffd@0 75 clear sm
wolffd@0 76 % As you can see, the smaller the gaussian kernel is, the more peaks can be
wolffd@0 77 % found in the novelty score. Indeed, if the kernel is small, the cumulative
wolffd@0 78 % multiplication of its elements with the superposed elements in the
wolffd@0 79 % similarity matrix may vary more easily, throughout the progressive
wolffd@0 80 % sliding of the kernel along the diagonal of the similarity matrix, and
wolffd@0 81 % local change of texture may be more easily detected. On the contrary,
wolffd@0 82 % when the kernel is large, only large-scale change of texture are
wolffd@0 83 % detected.
wolffd@0 84
wolffd@0 85 display('Strike any key to continue...');
wolffd@0 86 pause
wolffd@0 87 close all
wolffd@0 88
wolffd@0 89 p100 = mirpeaks(n100,'NoBegin','NoEnd')
wolffd@0 90 clear n100
wolffd@0 91 p50 = mirpeaks(n50,'NoBegin','NoEnd')
wolffd@0 92 clear n50
wolffd@0 93 p10 = mirpeaks(n10,'NoBegin','NoEnd')
wolffd@0 94 clear n10
wolffd@0 95 s100 = mirsegment(a,p100)
wolffd@0 96 clear p100
wolffd@0 97 mirplay(s100)
wolffd@0 98 clear s100
wolffd@0 99 s50 = mirsegment(a,p50)
wolffd@0 100 clear p50
wolffd@0 101 mirplay(s50)
wolffd@0 102 clear s50
wolffd@0 103 s10 = mirsegment(a,p10)
wolffd@0 104 clear p10
wolffd@0 105 mirplay(s10)
wolffd@0 106 clear s10
wolffd@0 107
wolffd@0 108 display('Strike any key to continue...');
wolffd@0 109 pause
wolffd@0 110 close all
wolffd@0 111
wolffd@0 112 % One more compact way of writing these commands is as follows:
wolffd@0 113 mirsegment(a,'Novelty')
wolffd@0 114 mirsegment(a,'Novelty','Contrast',0.01)
wolffd@0 115 mirsegment(a,'Novelty','KernelSize',100)
wolffd@0 116
wolffd@0 117 display('Strike any key to continue...');
wolffd@0 118 pause
wolffd@0 119 close all
wolffd@0 120
wolffd@0 121 % Besides, if you want to see the novelty curve with the peaks, just add a
wolffd@0 122 % second output:
wolffd@0 123 [s50 p50] = mirsegment(a,'Novelty','KernelSize',50)
wolffd@0 124 clear s50 p50
wolffd@0 125 [s10 p10] = mirsegment(a,'Novelty','KernelSize',10)
wolffd@0 126 clear a s10 p10
wolffd@0 127
wolffd@0 128 display('Strike any key to continue...');
wolffd@0 129 pause
wolffd@0 130 close all
wolffd@0 131
wolffd@0 132
wolffd@0 133 % 11. Try the whole process with MFCC instead of spectrum analysis. Take the
wolffd@0 134 % first ten MFCC for instance.
wolffd@0 135 help mirsegment
wolffd@0 136 % The segment function can simply be called as follows:
wolffd@0 137 sc = mirsegment('czardas','Novelty','MFCC','Rank',1:10)
wolffd@0 138 clear sc
wolffd@0 139
wolffd@0 140 % Here are some other examples of use:
wolffd@0 141 [ssp p m b] = mirsegment('valse_triste_happy','Spectrum',...
wolffd@0 142 'KernelSize',150,'Contrast',.1)
wolffd@0 143 clear p m b
wolffd@0 144 mirplay(ssp)
wolffd@0 145 clear ssp
wolffd@0 146
wolffd@0 147 display('Strike any key to continue...');
wolffd@0 148 pause
wolffd@0 149 close all
wolffd@0 150
wolffd@0 151 [smfcc2 p m a] = mirsegment('valse_triste_happy','MFCC',2:10,...
wolffd@0 152 'KernelSize',150,'Contrast',.1)
wolffd@0 153 clear p m a
wolffd@0 154 mirplay(smfcc2)