speech recognition

i have the following code for speech recognition using lpc parameters but its not working correctly:

b = wavrecord(Nseconds*Fs, Fs, 'double');
P = n1;             % Prediction order
P1 = P+1;

q = length(b);  
    % Length of the input speech
N = 250;            
% Length of each segment is 250

betold = zeros(1,P1);   % Set to zero the initial reverse prediction errors
for seg = 1:q/N,    % Loop whole speech
    s = b((seg-1)*N+1:seg*N);   % s is for one particular segment
    r = zeros(1,P1);    % Set to zero the initial autocorrelation
    for m = 1:P1,       % Loop to compute autocorrelation
       for n = 1:N-m+1,
            r(m) = r(m) + s(n)*s(n+m-1);
       end
    end                 % End of the autocorrelation loop
    a(1) = 0;           % Beginning of Durbin's recursive method
    error = r(1);
    for p = 1:P,
        aold = a;
          q = r(p+1);
          for m=1:p-1,
            q = q - a(m)*r(p-m+1);
          end
            if(error==0)
                error=1;
            end
          K(p) = q/error;        % Reflection coefficients
          error = error*(1.- K(p)*K(p));
          a(p) = K(p);
          for m=1:p-1,
            a(m) = aold(m) - K(p)*aold(p-m);
          end
    end
  f=q;
  fw(i,:) = f;
  display(fw);

[Walter Roberson]

Please edit your question, select your code and click on the 'Code {}' button. That will reformat your code so that it appears like a program to the readers.

[Walter Roberson]

I think you should expand on what you mean by "its not working correctly".

[Prince Ankit Singh]

LPC is a technique for compression of data in speech processing. Then hw can we implement it as a speech recognition technique?

Answer 1

[devika asokan]

showing a divide by 0 warning and coefficients all coming nearly equal.also the spoken word is not being correctly recognized.

 q = length(b); % Length of the input speech
 N = 250; 
% Length of each segment is 250
    betold = zeros(1,P1);
 % Set to zero the initial reverse prediction errors
 for seg = 1:q/N, % Loop whole speech
 s = b((seg-1)*N+1:seg*N); % s is for one particular segment
 r = zeros(1,P1); % Set to zero the initial autocorrelation 
for m = 1:P1,
 % Loop to compute autocorrelation 
for n = 1:N-m+1,
 r(m) = r(m) + s(n)*s(n+m-1);
 end
 end % End of the autocorrelation loop
 a(1) = 0; % Beginning of Durbin's recursive method 
error = r(1); 
for p = 1:P, 
aold = a; q = r(p+1); 
for m=1:p-1,
 q = q - a(m)*r(p-m+1); 
end 
 K(p) = q/error;
 % Reflection coefficients
 error = error*(1.- K(p)*K(p));
 a(p) = K(p); 
for m=1:p-1, 
a(m) = aold(m) - K(p)*aold(p-m); 
end 
end 
f=q; 
fw(i,:) = f;
 display(fw);

[Walter Roberson]

What would happen if the first sample or the last sample in the first segment were zero? What would the error become?