Can someone please help me with the code. Speech recognition system.

Hi there,

whenever I am trying to run the code, it first throws an error at "ms" in "mfcc.m" and then if cont. the running process, it stuck at the "x=disteu(......)" (line 25, "vector.m"). Can any one pls help with the code.

Phase one is main training code from where mfcc and vector code will be called.

[Walter Roberson]

what is the error message at ms and what are the sizes of the variables on that line ?

[Gaurav Sharma]

>> firstly it was showing an error regarding the index eceeding the array in mfcc.m . But after I made the change it stopped shwing that error. But now it is not showing any DCT plot.

"ms = MelFrequencyVector*abs(freqDomMat(1:nby2)).^2" => I change (1:nby2,:) to (1:nby2).

>> it enters into an infinite loop at "x= disteu(.....)"; line 25 => in the vector.m.

screen shot of workspace is attched down below for your kind reference.

[Walter Roberson]

if (((distortion - t)/t) < e) 

appears to be attempting to calculate a proportional error, but looks to be failing to take into account absolute value, that the range -e to +e should probably be accepted. Still, though, that would tend to break more often rather than less often.

[Gaurav Sharma]

But Sir, May I request you to tell me why I am not getting the DCT plot. And, for the above problem, What should I do ?

[Walter Roberson]

You have

ms = MelFrequencyVector*abs(freqDomMat(1:nby2)).^2; % mel spectrum
c = dct(log(ms));                                % mel-frequency cepstrum coefficients
c(1,:) = [] ;                                  % exclude 0'th order cepstral coefficient

Your ms is a row vector, 1 x 501. dct of a row vector gives you a row vector. You then ask to delete row 1 of the row vector, which is all that exists in the row vector, so your c is going to come out empty.

[Gaurav Sharma]

Thank You very mech sir for your help.

I need one more favour from you..... Can you pls tell why in the vector.m, the code enters in an infinite loop? and how it can be overcomed. I have tried but it didnt work.

[Walter Roberson]

In my test, if the data you pass as the first parameter of vector() has 1 or 2 columns, then the way you handle the means makes it certain that t will become 0. When it does then

(distortion - t)/t

has a division by 0 and becomes infinite, which is not less than e. This leads to 0 being assigned to distortion, and t becomes 0 again, leading to (0-0)/0 as the next test, which is NaN, which again is not less than e. You are then stuck in an infinite loop.

In the case of 3 or more columns then this problem is not certain to happen, and vector can potentially exit.