ivan@161: function reconstructed = AudioDeclipping_reconstruct(y, Problem)
ivan@136: %%  Audio declipping Problem reconstruction function
ivan@136: %   
ivan@136: %   This reconstruction function is using sparse representation y 
ivan@140: %   in dictionary Problem.A to reconstruct declipped audio.
ivan@161: %   The output structure has following fields:
ivan@161: %       audioAllSamples  - signal with all samples taken from reconstructed
ivan@161: %                          signal
ivan@161: %       audioOnlyClipped - only clipped samples are reconstructed,
ivan@161: %                          others are taken from original signal
ivan@161: %       snrAll           - psnr of whole signal
ivan@161: %       snrMiss          - psnr of the reconstructed clipped samples
ivan@140: %
ivan@140: %   [1] I. Damnjanovic, M. E. P. Davies, and M. P. Plumbley "SMALLbox - an 
ivan@140: %   evaluation framework for sparse representations and dictionary 
ivan@140: %   learning algorithms," V. Vigneron et al. (Eds.): LVA/ICA 2010, 
ivan@140: %   Springer-Verlag, Berlin, Germany, LNCS 6365, pp. 418-425
ivan@140: %   [2] A. Adler, V. Emiya, M. G. Jafari, M. Elad, R. Gribonval, and M. D.
ivan@140: %   Plumbley, “Audio Inpainting,” submitted to IEEE Trans. Audio, Speech,
ivan@140: %   and Lang. Proc., 2011, http://hal.inria.fr/inria-00577079/en/.
ivan@136: 
ivan@136: %
ivan@136: %   Centre for Digital Music, Queen Mary, University of London.
ivan@136: %   This file copyright 2009 Ivan Damnjanovic.
ivan@136: %
ivan@136: %   This program is free software; you can redistribute it and/or
ivan@136: %   modify it under the terms of the GNU General Public License as
ivan@136: %   published by the Free Software Foundation; either version 2 of the
ivan@136: %   License, or (at your option) any later version.  See the file
ivan@136: %   COPYING included with this distribution for more information.
ivan@136: %%
ivan@136: 
ivan@136: windowSize = Problem.windowSize;
ivan@136: overlap = Problem.overlap;
ivan@136: ws = Problem.ws(windowSize);
ivan@136: wa = Problem.wa(windowSize);
ivan@136: A = Problem.B;
ivan@136: 
ivan@136: orig     = Problem.original;
ivan@136: clipped  = Problem.clipped;
ivan@136: clipMask = Problem.clipMask;
ivan@136: 
ivan@136: % reconstruct audio frames
ivan@136: 
ivan@136: xFrames = diag(ws)*(A*y);
ivan@136: wNormFrames = (ws.*wa)'*ones(1,size(xFrames,2));
ivan@136: 
ivan@136: %   overlap and add
ivan@136: 
ivan@136: rec   = col2imstep(xFrames, size(clipped), [windowSize 1], [windowSize*overlap 1]);
ivan@136: wNorm = col2imstep(wNormFrames, size(clipped), [windowSize 1], [windowSize*overlap 1]); 
ivan@136: wNorm(find(wNorm==0)) = 1; 
ivan@136: recN  = rec./wNorm;
ivan@136: 
ivan@136: % change only clipped samples
ivan@136: 
ivan@136: recSignal = orig.*double(~clipMask) + recN.*double(clipMask); 
ivan@136: 
ivan@136: %% output structure image+psnr %%
ivan@136: reconstructed.audioAllSamples  = recN;
ivan@136: reconstructed.audioOnlyClipped = recSignal;
ivan@136: [reconstructed.snrAll , reconstructed.snrMiss] = SNRInpaintingPerformance(orig, clipped, recSignal, clipMask, 1);
ivan@136: 
ivan@136: end