Pre-processing-and-training.zip_For Real

function tree=buildTree(maleClusters,femaleClusters,UBMmale,UBMfemale,GMM,mfccFolder) %Organizes the database in a tree structure. %Inputs: % -(fe)maleClusters: cell array with arrays contaning speaker codes for % each cluster; % -UBMmale/UBMfemale is a struct with fields m(mean),v(variance) and % w(weight). v and m are k*n matrices, where k is the number of gaussians % used to build the UBM and n is the number of coefficients (in our case % this is equal to 39); % -GMM: cell array with GMM structs for each speaker. Each struct has % fields code(number of the speaker), gender(male or female), m(mean), % v(variance) and w(weight). Code is a number, gender is a string. % v and m are k*n matrices, where k is the number of gaussians used to % build the UBM and n is the number of coefficients (in our case this is % equal to 39); % -mfccFolder is the folder containing the MFCCs for each speaker in % separate .mat files. %Output: % -tree is a struct array, where each struct is a GMM with an additional % field named 'child', which contains another tree. [maleGMM,femaleGMM]=adaptClusters(maleClusters,femaleClusters,UBMmale,UBMfemale,mfccFolder); numClust = numel(maleClusters); for i = 1:numel(numClust) numComp = length(maleClusters{i}); for j = 1:numComp for k = 1:numel(GMM) if ((GMM{k}.code == maleClusters{i}(j)) && ... strcmp(GMM{k}.gender,'male')) child(j) = GMM{k}; end end end maleGMM{i}.child = child; maleData(i) = maleGMM{i}; end numClust = numel(femaleClusters); for i = 1:numel(numClust) numComp = length(femaleClusters{i}); for j = 1:numComp for k = 1:numel(GMM) if ((GMM{k}.code == femaleClusters{i}(j)) && ... strcmp(GMM{k}.gender,'female')) child(j) = GMM{k}; end end end femaleGMM{i}.child = child; femaleData(i) = femaleGMM{i}; end UBMmale.child = maleData; UBMfemale.child = femaleData; tree = [UBMmale,UBMfemale];