mgcOpt=mgcOptSet;
corpusDir='d:/dataSet/musicGenreClassification/GTZAN';

auSet=recursiveFileList(corpusDir, 'au');
auNum=length(auSet);
outputName=unique({auSet.parentDir});
for i=1:auNum
	auFile=auSet(i).path;
	fprintf('%d/%d: auFile=%s\n', i, auNum, auFile);
	wObj=waveFile2obj(auFile);
%	y=y*2^nbits/2;
%	temp=wave2mfccmex(y, fs, nbits, 'mfcc39.cfg');
%	auSet(i).mfcc=temp(1:13, :);
	mfccOpt=mfccOptSet(wObj.fs);
	mfccOpt.frameSize=1024;
	mfccOpt.overlap=512;
	mfccOpt.useDelta=2;
	auSet(i).mfcc=wave2mfcc(wObj.signal, wObj.fs, mfccOpt);
	auSet(i).output=find(strcmp(outputName, auSet(i).parentDir));
end

%fprintf('Saving auSet...\n');
%save auSet auSet

% ====== Feature extraction
for i=1:auNum
	fea=[];
	fea(end+1,:)=mean(auSet(i).mfcc, 2);	% Mean
	fea(end+1,:)=var(auSet(i).mfcc, 1, 2);	% Standard deviation
	fea(end+1,:)=min(auSet(i).mfcc, [], 2);	% Min
	fea(end+1,:)=max(auSet(i).mfcc, [], 2);	% Max
%	fea(end+1,:)=mean(abs(diff(auSet(i).mfcc, 1, 2)), 2);		% Mean of abs. diff
%	fea(end+1,:)=var(abs(diff(auSet(i).mfcc, 1, 2)), 2);		% Std of abs. diff
	auSet(i).fea=fea(:);
end

% ====== Create DS
DS.input=[auSet.fea];
%DS.input(end,:)=[];			% Why this input value is all the same?
DS.output=[auSet.output];
DS.outputName=outputName;

fprintf('Saving DS...\n');
save DS DS

fprintf('Loading DS.mat...\n');
load DS.mat
if mgcOpt.useInputNormalize, DS.input=inputNormalize(DS.input);	end		% Input normalization
%if mgcOpt.useInputWhiten, DS.input=inputWhiten(DS.input); end			% Input whitening
cvPrm.foldNum=mgcOpt.foldNum;
cvPrm.classifier=mgcOpt.classifier;
plotOpt=1;
figure; [tRrMean, vRrMean, tRr, vRr, computedClass]=crossValidate(DS, cvPrm, plotOpt);
%% === Confusion matrix
for i=1:length(computedClass)
	computed(i)=computedClass{i};
end
desired=DS.output;
confMat = confMatGet(desired, computed);
cmOpt=confMatPlot('defaultOpt');
confMatPlot(confMat, cmOpt);