Dawn@4: function [ idx ] = voicingByClustering( x, fs, noOfFrames, frameLength )
Dawn@4: % attempt to classify the voiced/unvoiced frames using k-means
Dawn@4: % clustering with the short term energy and spectral centroid
Dawn@4: % as feature vectors
Dawn@4: % returns an array containing voicing decision for frames
Dawn@4: % Useful only for speech frames
Dawn@4: % Window length and step (in seconds):
Dawn@4: win = frameLength/fs;
Dawn@4: step = win;
Dawn@4: 
Dawn@4: % calculate the short term energy
Dawn@4: Eor = ShortTimeEnergy(x, win*fs, step*fs );
Dawn@4: % calculate the spectral centroid 
Dawn@4: Cor = SpectralCentroid(x, win*fs, step*fs, fs );
Dawn@4: 
Dawn@4: % dataFileName = '../../../../../Results/VUVgrouping.txt';
Dawn@4: % dataFileID = fopen( dataFileName, 'w' );
Dawn@4: 
Dawn@4: noOfClusters = 2; %voiced, unvoiced
Dawn@4: data = [Eor Cor];
Dawn@4: idArray = zeros(1, length(Eor));
Dawn@4: 
Dawn@4: % myColours = ['r.'; 'm.'; 'c.'; 'w.'; 'g.'; 'y.'; 'b.'];
Dawn@4: 
Dawn@4: [idx ctrs]=kmeans( data, noOfClusters, 'Replicates',100,...
Dawn@4:     'start', 'sample', 'Distance', 'cityblock');
Dawn@4: 
Dawn@4: % we don't know which group will be classed as voiced
Dawn@4: % or unvoiced.
Dawn@4: % assume that the number of voiced frames is more than unvoiced
Dawn@4: 
Dawn@4: noOfAFrames = length( find( idx == 1 ));
Dawn@4: noOfBFrames = length( find( idx == 2 ));
Dawn@4: if( noOfAFrames < noOfBFrames )
Dawn@4:     voicedIdx = 2;
Dawn@4:     unvoicedIdx = 1;
Dawn@4: else
Dawn@4:     voicedIdx = 1;
Dawn@4:     unvoicedIdx = 2;
Dawn@4: end
Dawn@4: 
Dawn@4: 
Dawn@4: % now re-number the idx array so all voiced frames = 1 and
Dawn@4: % unvoiced = 2
Dawn@4: 
Dawn@4: voicedPos = find( idx == voicedIdx );
Dawn@4: unvoicedPos = find( idx == unvoicedIdx );
Dawn@4: 
Dawn@4: %replace the idx
Dawn@4: idx( voicedPos ) = 1;
Dawn@4: idx( unvoicedPos ) = 2;
Dawn@4: 
Dawn@4: end
Dawn@4: