ivan@161: %% DICTIONARY LEARNING FOR AUDIO DENOISING 
ivan@161: % This file contains an example of how SMALLbox can be used to test different
ivan@161: % dictionary learning techniques in Audio Denoising problem.
ivan@161: % It calls generateAudioDenoiseProblem that will let you to choose audio file,
ivan@161: % add noise and use noisy audio to generate training set for dictionary
ivan@161: % learning.
ivan@161: %   
ivan@161: %
ivan@161: %   Centre for Digital Music, Queen Mary, University of London.
ivan@161: %   This file copyright 2011 Ivan Damnjanovic.
ivan@161: %
ivan@161: %   This program is free software; you can redistribute it and/or
ivan@161: %   modify it under the terms of the GNU General Public License as
ivan@161: %   published by the Free Software Foundation; either version 2 of the
ivan@161: %   License, or (at your option) any later version.  See the file
ivan@161: %   COPYING included with this distribution for more information.
ivan@161: %   
ivan@161: %%
ivan@161: 
ivan@161: clear;
ivan@161: 
ivan@161: % Defining Audio Denoising Problem as Dictionary Learning
ivan@161: % Problem
ivan@161: 
ivan@161: SMALL.Problem = generateAudioDenoiseProblem('male01_8kHz',0.1,512,1/128,'','','',4);
ivan@161: 
ivan@161: %% 
ivan@161: %   Initialising solver structure
ivan@161: %   Setting solver structure fields (toolbox, name, param, solution,
ivan@161: %   reconstructed and time) to zero values
ivan@161: 
ivan@161: SMALL.solver(1)=SMALL_init_solver('MMbox', 'mm1', '', 1);
ivan@161: 
ivan@161: % Defining the parameters needed for image denoising
ivan@161: 
ivan@161: SMALL.solver(1).param=struct(...
ivan@161:     'lambda', 0.2,...
ivan@161:     'epsilon', 3*10^-4,...
ivan@161:     'iternum',10); 
ivan@161: 
ivan@161: %   Initialising Dictionary structure
ivan@161: %   Setting Dictionary structure fields (toolbox, name, param, D and time)
ivan@161: %   to zero values
ivan@161: 
ivan@161: SMALL.DL(1)=SMALL_init_DL('MMbox', 'MM_cn', '', 1);
ivan@161: 
ivan@161: 
ivan@161: %   Defining the parameters for MOD
ivan@161: %   In this example we are learning 256 atoms in 20 iterations, so that
ivan@161: %   every patch in the training set can be represented with target error in
ivan@161: %   L2-norm (EData)
ivan@161: %   Type help ksvd in MATLAB prompt for more options.
ivan@161: 
ivan@161: 
ivan@161: SMALL.DL(1).param=struct(...
ivan@161:     'solver', SMALL.solver(1),...
ivan@161:     'initdict', SMALL.Problem.initdict,...
ivan@161:     'dictsize', SMALL.Problem.p,...
ivan@161:     'iternum', 20,...
ivan@161:     'iterDictUpdate', 10,...
ivan@161:     'epsDictUpdate', 10^-7,...
ivan@161:     'cvset',0,...
ivan@161:     'show_dict', 0);
ivan@161: 
ivan@161: %   Learn the dictionary
ivan@161: 
ivan@161: SMALL.DL(1) = SMALL_learn(SMALL.Problem, SMALL.DL(1));
ivan@161: 
ivan@161: %   Set SMALL.Problem.A dictionary
ivan@161: %   (backward compatiblity with SPARCO: solver structure communicate
ivan@161: %   only with Problem structure, ie no direct communication between DL and
ivan@161: %   solver structures)
ivan@161: 
ivan@161: SMALL.Problem.A = SMALL.DL(1).D;
ivan@161: SMALL.Problem.reconstruct = @(x) AudioDenoise_reconstruct(x, SMALL.Problem);
ivan@161: %   Denoising the image - find the sparse solution in the learned
ivan@161: %   dictionary for all patches in the image and the end it uses
ivan@161: %   reconstruction function to reconstruct the patches and put them into a
ivan@161: %   denoised image
ivan@161: 
ivan@161: SMALL.solver(1)=SMALL_solve(SMALL.Problem, SMALL.solver(1));
ivan@161: 
ivan@161: %%
ivan@161: %%
ivan@161: % %  sparse coding using SPAMS online dictionary learning
ivan@161: % 
ivan@161: 
ivan@161: SMALL.DL(2)=SMALL_init_DL();
ivan@161: SMALL.DL(2).toolbox = 'SPAMS';
ivan@161: SMALL.DL(2).name = 'mexTrainDL';
ivan@161: SMALL.DL(2).param=struct('D', SMALL.Problem.initdict, 'K', SMALL.Problem.p, 'lambda', 0.2, 'iter', 200, 'mode', 3, 'modeD', 0);
ivan@161: 
ivan@161: 
ivan@161: SMALL.DL(2) = SMALL_learn(SMALL.Problem, SMALL.DL(2));
ivan@161: 
ivan@161: % Defining Reconstruction function
ivan@161: 
ivan@161: SMALL.Problem.A = SMALL.DL(2).D;
ivan@161: 
ivan@161: 
ivan@161: %%
ivan@161: % Initialising solver structure 
ivan@161: % Setting toolbox, name, param, solution, reconstructed and time to zero values
ivan@161: 
ivan@161: SMALL.solver(2)=SMALL_init_solver;
ivan@161: 
ivan@161: % Defining the parameters needed for sparse representation
ivan@161: 
ivan@161: SMALL.solver(2).toolbox='ompbox';
ivan@161: SMALL.solver(2).name='omp2';
ivan@161: SMALL.solver(2).param=struct(...
ivan@161:     'epsilon',0.2,...
ivan@161:     'maxatoms', 128); 
ivan@161: % Represent Training set in the learned dictionary 
ivan@161: 
ivan@161: SMALL.solver(2)=SMALL_solve(SMALL.Problem, SMALL.solver(2));
ivan@161: 
ivan@161: 
ivan@161: 
ivan@161: 
ivan@161: %%
ivan@161: % Plot results and save midi files
ivan@161: 
ivan@161: % show results %
ivan@161: 
ivan@161: 
ivan@161: SMALL_AudioDeNoiseResult(SMALL);
ivan@161: