How to do speech recognition using MFCC method in MATLAB?

Question

I'm developing a Speech Recognition engine for recognizing few (10-14) isolated words. I'm using MFCC (Mel Frequency Cepstral Coefficient) method and doing it using MATLAB. I'm referring a research paper and a website and other sources. I have mentioned the links below. Following are the main problem I'm facing:

1. Knowing whether the sampling frequency of 44100Hz that I have chosen is correct or not?

2. Whether the lower frequency=300Hz and upper frequency=8000Hz that is chosen to calculate Mel Filter Bank Matrix is correct or not?

3. Whether the frame-size=256, and no.of filter banks(coefficients)=20 that I have chosen is suitable for my application?

4. Whether pre-emphasis filter used is good? Or should I use butterworth filter?

5. Is normalisation of samples required before processing/ after filtering? Which normalisation technique should I use in that case?

CODE STRUCTURE:

RUN the MainRunning.m file. (Used to run code for the TRAINING SET once, and then the test samples)

SevenStep.m contains the code for training

SevenStepTestSample.m contains the code for testing

I have put my code in this link

Research Paper I'm referring is in this link

Reference website from where I studied calculation of Mel-Filter-Bank other than the research paper and other sites

Please suggest the error that I'm getting and the optimal upper frequency, lower frequency, frame-size, no. of filter banks for my application? How do I find the Hmatrix in my code correctly?(Mel Filter bank). Also do suggest any other errors, if any.

Thanks a lot in advance.

CODES:

MainRunning.m

    wordArray= {'begin', 'continue', 'exit', 'move', 'save', 'undo'};
% //NOT USING THESE REMAINING WORDS NOW :, 'a', 'b', 'c', 'd', 'one', 'two', 'three', 'four', 'a1', 'a2', 'a3', 'a4'};
SevenStep();
for j=1:6 %for all words in database
    ADDITION=0;
    disp(strcat('CHECKING FOR THE WORD: ',upper(wordArray{j})));
    for i=1:15 % each word has 15 samples for testing the accuracy of system
        [Value1]=SevenStepTestSample(strcat('C:\Users\Ray\Documents\BE Project\Phase-III\FreshAudioSamples\test\',wordArray{j},num2str(i),'.wav'));
        Result1=strcmpi(Value1,wordArray{j});
        %Result2=strcmpi(Value2,wordArray{j});
        ADDITION=ADDITION+((Result1));
    end
    fprintf('%s\t%d\n',upper(wordArray{j}),ADDITION); %Used to display no. of words correctly identified out of 15
end

SevenStep.m

function SevenStep()
    tic
    %-----------------------------------------------------------------INITIALISATION---------------------------------------------------------------%

    path_start='C:\Users\Ray\Documents\BE Project\Phase-III\FreshAudioSamples\';
    path_end='.wav';

    alpha=0.97;
    fs=44100; %frequency at which I have sampled my recorded samples
    pad_length=120064;
    numSamplesOfEachWord=30; %in database of samples
    numOfWords=6;
    frame_length=256; %How to choose an appropriate frame-size?
    no_of_frames=pad_length/(int32(frame_length/2)); %since overlap is 50%
    low_freq=300; %lower frequency for calculation of mel frequency filter bank (I'm unable to choose a correct one, and find the criteria for choosing it)
    high_freq=8000; %upper frequency for calculation of mel frequency filter bank (I'm unable to choose a correct one, and find the criteria for choosing it)
    no_of_coeffs=20;    % This is no. of Mel-Filter banks to create. how to choose a approriate value for this?

    %{

    fnm = 'C:\Users\Ray\Documents\BE Project\Phase-III\trifbank.mat'; 
    m1= matfile(fnm);
    H=m1.H;

    //Method 2 which I used to calculate when my Mel Frequency Filter Bank
    (given below) was probably not working correctly 

    %}

    %--------------------------------------------------PRE-PROCESSING FOR MEL FILTER BANK CREATION-----------------------------------------------%


    low_linear=2595*log10(1+(low_freq/700));
    high_linear=2595*log10(1+(high_freq/700));
    band_length=(high_linear-low_linear)/(no_of_coeffs+1); 

    MelArray(no_of_coeffs+2,1)=zeros(); %to store mel frequencies to calculate mel frequency filter bank
    LinearArray(no_of_coeffs+2,1)=zeros();  %to store linear frequencies to calculate mel frequency filter bank
    FreqArray(no_of_coeffs+2,1)=zeros();    %to store frequency array to calculate mel frequency filter bank
    %{
THIS ARRAY MAY HAVE WRONG VALUES DUE TO SELECTION of WRONG PARAMETERS LIKE low_freq, high_freq, frame_length (frame-size), no_of_coeffs (no. of filter banks). THIS IS MAJOR REASON BEHIND GENERATION OF NaN values in HMatrix
    %} 
    HMatrix(no_of_coeffs,frame_length)=zeros();  %Hmk Matrix/ Filter Bank
                                                 %I'M VERY DOUBTFUL OF THE
                                                 %VALUES GENERATED BY THIS
                                                 %FILTER BANK

    MelArray(1)=low_linear;
    MelArray(no_of_coeffs+2)=high_linear;
    LinearArray(1)=low_freq;
    LinearArray(no_of_coeffs+2)=high_freq;
    FreqArray(1)=floor((int32(frame_length)+1)*LinearArray(1)/fs);
    FreqArray(no_of_coeffs+2)=floor((int32(frame_length)+1)*LinearArray(no_of_coeffs+2)/fs);



    for m=1:no_of_coeffs
        MelArray(m+1)=MelArray(m)+band_length;
        LinearArray(m+1)=700*((power( 10,MelArray(m+1)/2595))-1);
        FreqArray(m+1)=floor((int32(frame_length)+1)*LinearArray(m+1)/fs);
    end


    % THE MOST DOUBTFUL PART i.e. MEL FREQUENCY FILTER BANK MATRIX CREATION

    %---------------------------------------------------------PROBABLE ERRONEOUS PART------------------------------------------------------------%
    % I'M GETTING NaN values in this matrix probably due to choosing
    % incorrect parameters for  like upper freq, lower freq, frame-size, no.of filter banks, sampling frequency etc.

    for k=1:frame_length
        for m=1:no_of_coeffs
            if(k<FreqArray(m))
                HMatrix(m,k)=0;
            elseif (FreqArray(m)<=k && k<=FreqArray(m+1))
                HMatrix(m,k)=(k-FreqArray(m))/(FreqArray(m+1)-FreqArray(m));
            elseif(FreqArray(m+1)<=k && k<=FreqArray(m+2))
                 HMatrix(m,k)=(FreqArray(m+2)-k)/(FreqArray(m+2)-FreqArray(m+1));
            elseif (k>FreqArray(m+2))
                 HMatrix(m,k)=0;
            end
        end
    end
    %--------------------------------------------------------------------------------------------------------------------------------------------%


    SM(no_of_frames,frame_length)=zeros();
    FVector(no_of_frames,no_of_coeffs,numOfWords*numSamplesOfEachWord)=zeros();
    LogFiltered(no_of_frames,no_of_coeffs)=zeros();
    FinalResult(no_of_frames,no_of_coeffs)=zeros();
    WindowValues=hamming(frame_length);

    wordArray= {'begin\', 'continue\', 'exit\', 'move\', 'save\', 'undo\'};
    % //NOT USING THESE REMAINING WORDS NOW : 'a\', 'b\', 'c\', 'd\', 'one\', 'two\', 'three\', 'four\', 'move\a1\', 'move\a2\', 'move\a3\', 'move\a4\'};

    %{
    //NOT USING THIS NOW
    filenaming = 'C:\Users\Ray\Documents\BE Project\Phase-III\Normalising.mat';
    m = matfile(filenaming);
    Nrm=m.Nrm;
    %} 

    %------------------------------------------------------------------TRAINING--------------------------------------------------------------------% 



    for i=1:numOfWords
        for j=1:numSamplesOfEachWord

            %----------------------------------------------------------READING WORD SAMPLE---------------------------------------------------%

           SD=wavread(strcat(path_start,wordArray{i},num2str(j),path_end));
           %SD=Nrm(j+(i-1)*numSamplesOfEachWord,:);     //NOT USING THIS NOW
           %SD=NoiseFilter(SD);                         % //NOT USING THIS NOW

           %----------------------------------------------------PRE-EMPHASIS FILTER-------------------------------------------------------%
           SD=filter([1 -alpha], 1, SD);               
           %------------------------------------------------------------PADDING-----------------------------------------------------------%

           SD=pad(SD,pad_length); % function given at the end of this file

           %-------------------------------------------------------FRAMING AND WINDOWING-------------------------------------------------------%
           %----------------------------------------------------AND CALCULATING FFT OF OBATINED SAMPLES--------------------------------------------%
           SD=reshape(SD,no_of_frames,frame_length/2);

           for k=1:no_of_frames-1
              temp=fft(([SD(k,:) SD(k+1,:)]'.*WindowValues)');
              SM(k,:)=temp.*(conj(temp));

           end
           temp=fft(([SD(no_of_frames,:) zeros(1,frame_length/2)]'.*WindowValues)');
           SM(no_of_frames,:)=temp.*(conj(temp));

           %-------------------------------------------------APPLYING MEL - FILTER BANK AND LOG FILTERING--------------------------------------------%
           for h=1:no_of_frames
                for l=1:no_of_coeffs
                    LogFiltered(h,l) = log10(sum(SM(h,:).*HMatrix(l,:)));
                end
           end

           %----------------------------------------------------------APPLYING DCT---------------------------------------------------------%           

           for indexing=1:no_of_frames
              FinalResult(indexing,:)=dct(LogFiltered(indexing,:));
           end

           %---------------------------------------------------STORING TRAINING DATASET---------------------------------------------------%

           FVector(1:no_of_frames,1:no_of_coeffs,j+(i-1)*numSamplesOfEachWord)=FinalResult;

        end
    end

    save('SevenStep');   %Saving all variables to retrieve during running the program for SR
    toc
end

function p=pad(x,full_length)
    p = vertcat(x, zeros(full_length - length(x),1));
end


%{
NOT USING THIS NOW, BUT IF YOU CAN SUGGEST CHANGES IN THIS IT WOULD
APPRECIABLE

function Y=NoiseFilter(SD)
    [B, A]= butter(2,0.01); % IS THIS CORRECT FOR MY SAMPLES? 
    X= filter(B, A, SD);
    %figure; plot(SD);
    X(abs(X)< 0.01) = 0;
    %figure; plot(X);
    %X=X(find(X,1,'first'):find(X,1,'last'));
    Y=SD(find(X,1,'first'):find(X,1,'last')); % TAKING FROM ORIGINAL SAMPLE
    %figure; plot(Y);
end
%} 

SevenStepTestSample.m

function [StrVal1] =SevenStepTestSample(filename)

   tic

    alpha=0.97;
    KVal=13; %The no. of values used for final comparison
    pad_length=120064;
    numSamplesOfEachWord=10;
    numOfWords=6;
    frame_length=256; %How to choose an appropriate frame-size?
    no_of_frames=pad_length/(frame_length/2); %since overlap is 50%
    no_of_coeffs=20;    % This is no. of Mel-Filter banks to create. how to choose a approriate value for this?

    wordArrayPrint= {'begin', 'continue', 'exit', 'move', 'save', 'undo'};
    % NOT USING THE REMAINING WORDS FOR NOW :'a', 'b', 'c', 'd', 'one', 'two', 'three', 'four', 'a1', 'a2', 'a3', 'a4'};   

    filenm = 'C:\Users\Ray\Documents\BE Project\Phase-III\SevenStep.mat'; % This is the storage space of training set data
    m = matfile(filenm);
    FVector=m.FVector; 
    HMatrix=m.HMatrix;
    WindowValues=m.WindowValues;

    TSM(no_of_frames,frame_length)=zeros();
    TLogFiltered(no_of_frames,no_of_coeffs)=zeros();
    TFinalResult(no_of_frames,no_of_coeffs)=zeros();

    [testSample,t1] = wavread(filename);

    %{
    WAS USING THIS FOR NORMALISATION OF VOICE SAMPLES, NOT USING IT NOW

    tt=max(abs(testSample));
    for j=1:length(testSample)
    RD(j)=(testSample(j)/tt);
    end
    testSample=RD;
    %}


    % NOT USING THIS NOW : testSample=NoiseFilter(testSample);
    testSample=filter([1 -alpha],1,testSample');
    testSample = vertcat(testSample', zeros(pad_length - length(testSample),1)); 
    testSample=reshape(testSample,no_of_frames,frame_length/2);

    for k=1:no_of_frames-1
        temp=fft(([testSample(k,:) testSample(k+1,:)]'.*WindowValues)');
        TSM(k,:)=temp.*(conj(temp));
    end
    temp=fft(([testSample(no_of_frames,:) zeros(1,frame_length/2)]'.*WindowValues)');
    TSM(no_of_frames,:)=temp.*(conj(temp));

    for h=1:no_of_frames
        for l=1:no_of_coeffs
            TLogFiltered(h,l) = log10(sum(TSM(h,:).*HMatrix(l,:)));
        end
    end

    for indexing=1:no_of_frames
        TFinalResult(indexing,:)=dct(TLogFiltered(indexing,:));
    end


    errorArray = zeros(numOfWords*numSamplesOfEachWord,1);

    for j=1:numOfWords*numSamplesOfEachWord
           errorArray(j)= mean2(abs(TFinalResult(1:no_of_frames,2:KVal) - FVector(1:no_of_frames,2:KVal,j)));
    end     

    %errorArray'


   [err, index1] = min(errorArray);

   %{
   NOT USING THIS NOW: THIS CODE IS TO CALCULATE SECOND MINIMUM ASSUMING
   THE FACT THAT NO TWO SAMPLES WILL GIVE SAME ERROR VALUE
   [err,index2] = min(errorArray(errorArray>min(errorArray)));
    if(index1<=index2)
       index2=index2+1;
     end
   %} 
   for j=1:numOfWords
    if (index1 >= (1+(j-1)*numSamplesOfEachWord) && index1 <= j*numSamplesOfEachWord)
        StrVal1=upper(wordArrayPrint{j});
        disp(StrVal1);
    end

    %{
        if (index2 >= (1+(j-1)*numSamplesOfEachWord) && index2 <= j*numSamplesOfEachWord)
            StrVal2=upper(wordArrayPrint{j});
        end
    %}
   end

       %fprintf('%s\t%s\n',StrVal1,StrVal2);
       % save('SevenStepTestSample');

  toc

end