annotate Code/Classifiers/kmeans_MFCC_Singing.m @ 4:92ca03a8fa99 tip

Update to ICASSP 2013 benchmark

function [] = kmeans_MFCC_Singing( inputFileName, outputFileName )

cd 'C:\Users\dawn\Dropbox\TestResults'
inputFileName

DEBUG = 1;
% output results file name
masterFileOutputID = fopen( outputFileName, 'a' );   %fopen( 'kmeans_Singing_MFCC.txt', 'a' );
% % input results file name
% inputFileName = 'singingMFCCStats_VoicedAndUnvoiced.txt';

fprintf( masterFileOutputID, '\n RESULTS FILE NAME: %s\n', inputFileName);
inputFileID = fopen( inputFileName );

% noOfArguments = length(varargin);
% 
% outputFileName = 'individualResults/kmeans_Results_';
% resultsFileName = 'kmeans_Results_';
titleName = '';
% for i=1 : noOfArguments
%     titleName = [ titleName varargin{i} '_'];
%     fprintf( masterFileOutputID, '%s_', varargin{i} );
% end
% 
% outputFileName = [ outputFileName titleName ];
% resultsFileName = [ resultsFileName titleName ];

fprintf( masterFileOutputID, '\t' );

% outputFileName = [ outputFileName '.txt'];
% resultsFileName = [ resultsFileName '.txt'];

% fileOutputID = fopen( outputFileName, 'w' );
% fileKMeansOutputID = fopen( resultsFileName, 'w' );

% -------------------- get the data from the results file ---------------
lineCount = 0;
fileCount = 0;
data = [];
while( ~(feof(inputFileID)) )
    
    outputValues = [];
    thestr = fgetl(inputFileID);
    fileCount = fileCount + 1;

    % determine whether we have a positive or negative sample
    sampleEmotion( fileCount ) = 'U';
    if( ~(isempty(strfind(thestr,'pos'))))
        % sample is positive
        sampleEmotion( fileCount ) = 'P';
    elseif( ~(isempty(strfind(thestr,'neg'))))
        % sample is negative
        sampleEmotion( fileCount ) = 'N';
    else
        disp('EEEK!');
        pause;
    end

    % determine whether we have a male, female or trans sample
    gender( fileCount ) = '?';
    if( ~(isempty(strfind(thestr,'fem'))))
        % gender is female
        gender( fileCount ) = 'F';
    elseif( ~(isempty(strfind(thestr,'male'))))
        % gender is male
        gender( fileCount ) = 'M';
    elseif( ~(isempty(strfind(thestr,'trans'))))
        % gender is trans
        gender( fileCount ) = 'T';
    else
        disp('EEEK!');
        pause;
    end

    %how many values are in the string?
    spaces = strfind( thestr, ' ' );
    numberstr = thestr( spaces(1) : end ); % chop off the file name  
    vars = sscanf( numberstr, '%f', inf );
    data( fileCount, : ) = vars;
        
    lineCount = lineCount + 1;
    
end
fclose(inputFileID);

% try with and without different metrics classes
% data is a 162 variable array of MFCC stats
% first there are 6 metrics for each of the 13 frequency bands (mean, median, var, min, max, range) 
% then there are 6 metrics for the 1st derivative of the 13 frequency bands (mean, median, var, min, max, range) 
% then there are 6 metrics for the mean of all frequency bands

%  dataOptions = (1:1:(13*12)+6); % fprintf( masterFileOutputID, '\n try with all \n');
% dataOptions = (1:1:13*6);  fprintf( masterFileOutputID, '\n try with just the ordinary 13 frequency bands \n');
% dataOptions = (1:1:13*6);  fprintf( masterFileOutputID, '\n try with just the ordinary 13 frequency bands and without the median or range \n');

% ------------  apply the k-means classifier  ------------------------

noOfClusters = 2; % we are only trying to identify positive and negative emotions


[idx ctrs]=kmeans( data, noOfClusters, 'Replicates',100,...
    'start', 'sample', 'Distance', 'cityblock');

%display results grouped by emotion
fprintf( masterFileOutputID, '\n Emotion grouping \n');
fprintf( masterFileOutputID, 'cityblock \n');
[ groupStats, groupNames ] = processKMeansResults( 'cityblock', idx, sampleEmotion, masterFileOutputID, titleName, DEBUG );
[ confusionMatrix ] = getConfusionMatrix( groupStats, groupNames, masterFileOutputID );
% if(DEBUG == 1)
%     disp('press space');
%     pause;
% end

fprintf( masterFileOutputID, 'sqEuclidean \n');
[idx ctrs]=kmeans( data, noOfClusters, 'Replicates',100,...
    'start', 'sample', 'Distance', 'sqEuclidean');

[ groupStats, groupNames ] = processKMeansResults( 'sqEuclidean', idx, sampleEmotion, masterFileOutputID, titleName, DEBUG );
[ confusionMatrix ] = getConfusionMatrix( groupStats, groupNames, masterFileOutputID );

% if(DEBUG == 1)
%     disp('press space');
%     pause;
% end

% display results grouped by gender
fprintf( masterFileOutputID, '\n Gender grouping \n');
noOfClusters = 3;

[idx ctrs]=kmeans( data, noOfClusters, 'Replicates',100,...
    'start', 'sample', 'Distance', 'cityblock');

fprintf( masterFileOutputID, 'cityblock \n');
[ groupStats, groupNames ] = processKMeansResults( 'cityblock', idx, gender, masterFileOutputID, titleName, DEBUG );
[ confusionMatrix ] = getConfusionMatrix( groupStats, groupNames, masterFileOutputID );

[idx ctrs]=kmeans( data, noOfClusters, 'Replicates',100,...
    'start', 'sample', 'Distance', 'sqEuclidean');

fprintf( masterFileOutputID, 'sqEuclidean \n');
[ groupStats, groupNames ] = processKMeansResults( 'sqEuclidean', idx, gender, masterFileOutputID, titleName, DEBUG );
[ confusionMatrix ] = getConfusionMatrix( groupStats, groupNames, masterFileOutputID );


% fprintf( fileOutputID, '\n' );
% fclose( fileOutputID );
% fprintf( fileKMeansOutputID, '\n' );
% fclose( fileKMeansOutputID );
fprintf( masterFileOutputID, '\n' );
fclose( masterFileOutputID );

end