Mercurial > hg > smallbox
view Problems/AMT_reconstruct.m @ 181:0dc98f1c60bb danieleb
minor edits
author | Daniele Barchiesi <daniele.barchiesi@eecs.qmul.ac.uk> |
---|---|
date | Thu, 05 Jan 2012 15:46:13 +0000 |
parents | f42aa8bcb82f |
children |
line wrap: on
line source
function reconstructed=AMT_reconstruct(V, Problem) %% Reconstruction of midi file from representation in the given dictionary % % SMALL_midiGenerate is a part of SMALLbox and can be use to reconstruct % a midi file given representation of the training set (V) in the % dictionary Problem.A. % Output is reconstructed structure with two fields: % - reconstructed.notes - matrix with transcribed notes % - reconstructed.midi - midi representation of transcription % % Centre for Digital Music, Queen Mary, University of London. % This file copyright 2009 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. %% U=Problem.A; % Dictionary used for representation fs=Problem.fs; % Sampling rate f=Problem.f; % vector of frequencies at wihch spectrogram is computed ts=(Problem.windowSize*(1-Problem.overlap))/fs; %size of an analysis frame in seconds %% % Components pitch estimation using modified SWIPE algorithm by Arthuro % Camacho % % Columns of matrix U are spectrograms of the notes learned from the % training set. We are estimating pitches of these notes by also % restricting pitch values to the one of the 88 piano notes. pitch=zeros(size(U,2),1); for i=1:size(U,2) pitch(i) = SMALL_swipe(U(:,i),fs, f, [27.50 8192], 1/12); end %% % If some of columns of U have the same pitch, their contribution to the % score (matrix V) is summed. [Ps,idx]=sort(pitch); ndp=1; Pd(ndp)=Ps(1); Vnew(ndp,:)=V(idx(1),:); for i=2:88 if Ps(i)> Ps(i-1) ndp=ndp+1; Vnew(ndp,:)=V(idx(i),:); Pd(ndp)=Ps(i); else Vnew(ndp,:)=Vnew(ndp,:)+V(idx(i),:); end end %% % Generate midi matrix midx=0; for i=1:ndp % Threshold for finding onsets and offsets of notes thr=mean(Vnew(i,:));%+std(Vnew(i,:)); if(Pd(i)~=0) for j=1:size(Vnew,2) if Vnew(i,j)<thr Vnew(i,j)=0; if(j>1) if (Vnew(i,j-1)==1) try M(midx,6)=(j-1)*ts; if (M(midx,6)-M(midx,5))<2*ts midx=midx-1; end catch pause; end end end else Vnew(i,j)=1; if(j>1) if (Vnew(i,j-1)==0) midx=midx+1; M(midx,1)=1; M(midx,2)=1; M(midx,3)=69 +round( 12 *log2(Pd(i)/440)); M(midx,4)=80; M(midx,5)=(j-1)*ts; end else midx=midx+1; M(midx,1)=1; M(midx,2)=1; M(midx,3)=69 + round(12 *log2(Pd(i)/440)); M(midx,4)=80; M(midx,5)=0; end end end if M(midx,6)==0 M(midx,6)=(j-1)*ts; end end end M=sortrows(M,5); reconstructed.notes=M; reconstructed.midi = matrix2midi(M);