annotate examples/Automatic Music Transcription/SMALL_AMT_KSVD_Err_test.m @ 128:8e660fd14774 ivand_dev

Feature 186
author Ivan Damnjanovic lnx <ivan.damnjanovic@eecs.qmul.ac.uk>
date Mon, 13 Jun 2011 14:55:45 +0100
parents cbf3521c25eb
children f42aa8bcb82f
rev   line source
ivan@128 1 %% Dictionary Learning for Automatic Music Transcription - KSVD residual
ivan@128 2 %% error test
ivan@128 3 %
ivan@128 4 % *WARNING!* You should have SPAMS in your search path in order for this
ivan@128 5 % script to work.Due to licensing issues SPAMS can not be automatically
ivan@128 6 % provided in SMALLbox (http://www.di.ens.fr/willow/SPAMS/downloads.html).
idamnjanovic@25 7 %
idamnjanovic@6 8 % This file contains an example of how SMALLbox can be used to test diferent
idamnjanovic@6 9 % dictionary learning techniques in Automatic Music Transcription problem.
idamnjanovic@6 10 % It calls generateAMT_Learning_Problem that will let you to choose midi,
idamnjanovic@6 11 % wave or mat file to be transcribe. If file is midi it will be first
idamnjanovic@6 12 % converted to wave and original midi file will be used for comparison with
idamnjanovic@6 13 % results of dictionary learning and reconstruction.
idamnjanovic@6 14 % The function will generarte the Problem structure that is used to learn
idamnjanovic@6 15 % Problem.p notes spectrograms from training set Problem.b using
idamnjanovic@6 16 % dictionary learning technique defined in DL structure.
idamnjanovic@6 17 %
idamnjanovic@25 18
ivan@128 19 %
ivan@128 20 % Centre for Digital Music, Queen Mary, University of London.
ivan@128 21 % This file copyright 2010 Ivan Damnjanovic.
ivan@128 22 %
ivan@128 23 % This program is free software; you can redistribute it and/or
ivan@128 24 % modify it under the terms of the GNU General Public License as
ivan@128 25 % published by the Free Software Foundation; either version 2 of the
ivan@128 26 % License, or (at your option) any later version. See the file
ivan@128 27 % COPYING included with this distribution for more information.
idamnjanovic@6 28 %%
idamnjanovic@6 29
idamnjanovic@6 30 clear;
idamnjanovic@6 31
idamnjanovic@6 32
idamnjanovic@6 33 % Defining Automatic Transcription of Piano tune as Dictionary Learning
idamnjanovic@6 34 % Problem
idamnjanovic@6 35
idamnjanovic@6 36 SMALL.Problem = generateAMT_Learning_Problem();
idamnjanovic@6 37 TPmax=0;
idamnjanovic@6 38 for i=1:5
idamnjanovic@6 39 %%
idamnjanovic@6 40 % Use KSVD Dictionary Learning Algorithm to Learn 88 notes (defined in
idamnjanovic@6 41 % SMALL.Problem.p) using sparsity constrain only
idamnjanovic@6 42
idamnjanovic@6 43 % Initialising Dictionary structure
idamnjanovic@6 44 % Setting Dictionary structure fields (toolbox, name, param, D and time)
idamnjanovic@6 45 % to zero values
idamnjanovic@6 46
idamnjanovic@6 47 SMALL.DL(i)=SMALL_init_DL(i);
idamnjanovic@6 48
idamnjanovic@6 49 % Defining the parameters needed for dictionary learning
idamnjanovic@6 50
idamnjanovic@6 51 SMALL.DL(i).toolbox = 'KSVD';
idamnjanovic@6 52 SMALL.DL(i).name = 'ksvd';
idamnjanovic@6 53
idamnjanovic@6 54 % Defining the parameters for KSVD
idamnjanovic@6 55 % In this example we are learning 88 atoms in 100 iterations, so that
idamnjanovic@6 56 % every frame in the training set can be represented with maximum 10
idamnjanovic@6 57 % dictionary elements. However, our aim here is to show how individual
idamnjanovic@6 58 % parameters can be ested in the AMT problem. We test five different
idamnjanovic@6 59 % values for residual error (Edata) in KSVD algorithm.
idamnjanovic@6 60 % Type help ksvd in MATLAB prompt for more options.
idamnjanovic@6 61
idamnjanovic@6 62 Edata(i)=8+i*2;
idamnjanovic@6 63 SMALL.DL(i).param=struct(...
idamnjanovic@6 64 'Edata', Edata(i),...
idamnjanovic@6 65 'dictsize', SMALL.Problem.p,...
idamnjanovic@6 66 'iternum', 100,...
idamnjanovic@6 67 'maxatoms', 10);
idamnjanovic@6 68
idamnjanovic@6 69 % Learn the dictionary
idamnjanovic@6 70
idamnjanovic@6 71 SMALL.DL(i) = SMALL_learn(SMALL.Problem, SMALL.DL(i));
idamnjanovic@6 72
idamnjanovic@6 73 % Set SMALL.Problem.A dictionary and reconstruction function
idamnjanovic@6 74 % (backward compatiblity with SPARCO: solver structure communicate
idamnjanovic@6 75 % only with Problem structure, ie no direct communication between DL and
idamnjanovic@6 76 % solver structures)
idamnjanovic@6 77
idamnjanovic@6 78 SMALL.Problem.A = SMALL.DL(i).D;
idamnjanovic@6 79 SMALL.Problem.reconstruct = @(x) SMALL_midiGenerate(x, SMALL.Problem);
idamnjanovic@6 80
idamnjanovic@6 81 %%
idamnjanovic@6 82 % Initialising solver structure
idamnjanovic@6 83 % Setting solver structure fields (toolbox, name, param, solution,
idamnjanovic@6 84 % reconstructed and time) to zero values
idamnjanovic@6 85 % As an example, SPAMS (Julien Mairal 2009) implementation of LARS
idamnjanovic@6 86 % algorithm is used for representation of training set in the learned
idamnjanovic@6 87 % dictionary.
idamnjanovic@6 88
idamnjanovic@6 89 SMALL.solver(1)=SMALL_init_solver;
idamnjanovic@6 90
idamnjanovic@6 91 % Defining the parameters needed for sparse representation
idamnjanovic@6 92
idamnjanovic@6 93 SMALL.solver(1).toolbox='SPAMS';
idamnjanovic@6 94 SMALL.solver(1).name='mexLasso';
idamnjanovic@6 95 SMALL.solver(1).param=struct('lambda', 2, 'pos', 1, 'mode', 2);
idamnjanovic@6 96
idamnjanovic@6 97 %Represent Training set in the learned dictionary
idamnjanovic@6 98
idamnjanovic@6 99 SMALL.solver(1)=SMALL_solve(SMALL.Problem, SMALL.solver(1));
idamnjanovic@6 100
idamnjanovic@6 101 %%
idamnjanovic@6 102 % Analysis of the result of automatic music transcription. If groundtruth
idamnjanovic@6 103 % exists, we can compare transcribed notes and original and get usual
idamnjanovic@6 104 % True Positives, False Positives and False Negatives measures.
idamnjanovic@6 105
idamnjanovic@6 106 AMT_res(i) = AMT_analysis(SMALL.Problem, SMALL.solver(1));
idamnjanovic@6 107
idamnjanovic@6 108 if AMT_res(i).TP>TPmax
idamnjanovic@6 109 TPmax=AMT_res(i).TP;
idamnjanovic@6 110 BLmidi=SMALL.solver(1).reconstructed.midi;
idamnjanovic@6 111 max=i;
idamnjanovic@6 112 end
idamnjanovic@6 113
idamnjanovic@6 114 end %end of for loop
idamnjanovic@6 115
idamnjanovic@6 116 %%
idamnjanovic@6 117 % Plot results and save midi files
idamnjanovic@6 118
idamnjanovic@6 119 figAMTbest=SMALL_AMT_plot(SMALL, AMT_res(max));
idamnjanovic@6 120
idamnjanovic@6 121 resFig=figure('Name', 'Automatic Music Transcription KSVD Error TEST');
idamnjanovic@6 122
idamnjanovic@6 123 subplot (3,1,1); plot(Edata(:), [AMT_res(:).TP], 'ro-');
idamnjanovic@6 124 title('True Positives vs Edata');
idamnjanovic@6 125
idamnjanovic@6 126 subplot (3,1,2); plot(Edata(:), [AMT_res(:).FN], 'ro-');
idamnjanovic@6 127 title('False Negatives vs Edata');
idamnjanovic@6 128
idamnjanovic@6 129 subplot (3,1,3); plot(Edata(:), [AMT_res(:).FP], 'ro-');
idamnjanovic@6 130 title('False Positives vs Edata');
idamnjanovic@6 131
idamnjanovic@6 132 FS=filesep;
idamnjanovic@6 133 [pathstr1, name, ext, versn] = fileparts(which('SMALLboxSetup.m'));
idamnjanovic@6 134 cd([pathstr1,FS,'results']);
idamnjanovic@6 135 [filename,pathname] = uiputfile({' *.mid;' },'Save midi');
idamnjanovic@6 136 if filename~=0 writemidi(BLmidi, [pathname,FS,filename]);end
idamnjanovic@6 137 [filename,pathname] = uiputfile({' *.fig;' },'Save figure TP/FN/FP vs lambda');
idamnjanovic@6 138 if filename~=0 saveas(resFig, [pathname,FS,filename]);end
idamnjanovic@6 139
idamnjanovic@6 140 [filename,pathname] = uiputfile({' *.fig;' },'Save BEST AMT figure');
idamnjanovic@6 141 if filename~=0 saveas(figAMTbest, [pathname,FS,filename]);end
idamnjanovic@6 142