Mercurial > hg > smallbox
view Problems/generateAudioDenoiseProblem.m @ 164:4205744092e6 release_1.9
Merge from branch "ivand_dev"
author | Ivan Damnjanovic lnx <ivan.damnjanovic@eecs.qmul.ac.uk> |
---|---|
date | Wed, 07 Sep 2011 14:17:30 +0100 |
parents | f42aa8bcb82f |
children | 9c418bea7f6a |
line wrap: on
line source
function data = generateAudioDenoiseProblem(soundfile, sigma, windowSize,... overlap, wa, ws, trainnum, redundancyFactor, initdict) %% Audio Denoising Problem % % generateAudioDenoiseProblem is part of the SMALLbox and generate a % problem for comaprison of Dictionary Learning/Sparse Representation % techniques in audio denoising scenario. % % The function takes as an optional input % soundfile - name of the file % sigma - noise level (dB) % windowSize - 1D frame size (eg 512) % overlap - ammount of overlaping frames between 0 and 1 % wa,ws - analisys and synthesis window functions % % trainnum - number of frames for training % redundancyFactor - overcompletness of dictionary (default 2) % initdict - initial dictionary % % The function outputs the structure with following fields: % Original - original signal % Noisy - signal with added noise % fs - sample rate of the original signal in Hertz % nbits - the number of bits per sample % sigma - added noise level % b - matrix of training samples for dictionary learning % b1 - matrix containing all frames for reconstruction step % m - size od dictionary atom % n - number of frames for training % p - number of atoms in dictionary % windowSize - 1D frame size (eg 512) % overlap - ammount of overlaping frames between 0 and 1 % wa,ws - analisys and synthesis window functions % initdict - initial dictionary % Centre for Digital Music, Queen Mary, University of London. % This file copyright 2011 Ivan Damnjanovic. % % This program is free software; you can redistribute it and/or % modify it under the terms of the GNU General Public License as % published by the Free Software Foundation; either version 2 of the % License, or (at your option) any later version. See the file % COPYING included with this distribution for more information. %% disp(' '); disp(' ********** Denoising Problem **********'); disp(' '); disp(' This function reads an audio, adds random Gaussian noise,'); disp(' that can be later denoised by using dictionary learning techniques.'); disp(' '); FS=filesep; %% prompt user for wav file %% %ask for file name TMPpath=pwd; if ~ exist( 'soundfile', 'var' ) || isempty(soundfile) %ask for file name [pathstr1, name, ext, versn] = fileparts(which('SMALLboxSetup.m')); cd([pathstr1,FS,'data',FS,'audio']); [filename,pathname] = uigetfile({'*.mat; *.mid; *.wav'},'Select a file to transcribe'); [pathstr, name, ext, versn] = fileparts(filename); data.name=name; if strcmp(ext,'.mid') midi=readmidi(filename); % data.notesOriginal=midiInfo(midi); y=midi2audio(midi); wavwrite(y, 44100, 16, 'temp.wav'); [x.signal, x.fs, x.nbits]=wavread('temp.wav'); delete('temp.wav'); elseif strcmp(ext,'.wav') % cd([pathstr1,FS, 'data', FS, 'audio', FS, 'midi']); % filename1=[name, '.mid']; % if exist(filename1, 'file') % midi=readmidi(filename1); % data.notesOriginal=midiInfo(midi); % end cd([pathstr1,FS, 'data', FS, 'audio', FS, 'wav']); [x.signal, x.fs, x.nbits]=wavread(filename); else % cd([pathstr1,FS, 'data', FS, 'audio', FS, 'midi']); % filename1=[name, '.mid']; % if exist(filename1, 'file') % midi=readmidi(filename1); % data.notesOriginal=midiInfo(midi); % end cd([pathstr1,FS, 'data', FS, 'audio', FS, 'mat']); x=load([pathname,filename]); end else [x.signal, x.fs, x.nbits]=wavread(soundfile); [pathstr, name, ext, versn] = fileparts(soundfile); data.name=name; end %% set parameters %% if ~ exist( 'sigma', 'var' ) || isempty(sigma), sigma = 0.2; end if ~ exist( 'windowSize', 'var' ) || isempty(windowSize), windowSize = 256;end if ~ exist( 'overlap', 'var' ) || isempty(overlap), overlap = 0.5; end if ~ exist( 'wa', 'var' ) || isempty(wa), wa = @wSine; end % Analysis window if ~ exist( 'ws', 'var' ) || isempty(ws), ws = @wSine; end % Synthesis window if ~ exist( 'redundancyFactor', 'var' ) || isempty(windowSize),... redundancyFactor = 2;end if ~ exist( 'initdict', 'var' ) || isempty(initdict),... initdict = 'odct'; end if ~ exist( 'trainnum', 'var' ) || isempty(trainnum), ... trainnum = 16*redundancyFactor*windowSize;end if (strcmpi(initdict,'odct')) initdict = odctndict(windowSize, redundancyFactor*windowSize, 1); elseif (strcmpi(initdict,'data')) clear initdict; % causes initialization using random examples else error('Invalid initial dictionary specified.'); end if exist( 'initdict', 'var' ) initdict = initdict(:,1:redundancyFactor*windowSize); end %%%% create training data %%% %% generate noisy audio %% disp(' '); disp('Generating noisy audio...'); x.signal = x.signal/max(abs(x.signal(:)))*0.99999; n = randn(size(x.signal)) .* sigma; xnoise = x.signal + n;% here we can load noise audio if available % for example: wavread('icassp06_x.wav');% X = im2colstep(xnoise,[windowSize 1],[overlap*windowSize 1]); X = diag(wa(windowSize)) * X; % remove dc in blocks to conserve memory % % bsize = 2000; % for i = 1:bsize:size(X,2) % blockids = i : min(i+bsize-1,size(X,2)); % X(:,blockids) = remove_dc(X(:,blockids),'columns'); % end data.Original = x.signal; data.Noisy = xnoise; data.fs = x.fs; data.nbits = x.nbits; data.sigma = sigma; if (trainnum<size(X,2)) p = randperm(size(X,2)); p=sort(p(1:trainnum)); data.b = X(:,p); else data.b = X; end data.b1 = X; [data.m, data.n] = size(data.b); data.p = redundancyFactor*windowSize; data.windowSize = windowSize; data.overlap = overlap; data.ws = ws; data.wa = wa; data.initdict= initdict; cd(TMPpath);