Voice Recognition - how to add a threshold value ?

Dear friends,

i have used a problem with a matlab project which is to identify voice and display numbers between 0-9 using Linear Predictive Code (LPC) approach. The recorded voices are stored in a database and using a neural network the training process is carried using those LPC coefficients are stored in the database. Do i have to use a threshold value for this problem ?

Ex - when a user speak zero it display 0 , five -> 5. The code is given below, the problem i am facing now is that even if i say a word like EAT it displays a number which should not happen. What is the solution for this and to display NOT FOUND? i really appreciate if someone can help me in this.

Code :-

if true
clc;
clear all;
load('voicetrainfinal.mat');
Fs=8000;
for l=1:20
clear y1 y2 y3;
display('Press ENTER to record your voice !');
pause();
x=wavrecord(Fs,Fs);
t=0.04;
j=1;
for i=1:8000
  if(abs(x(i))>t)
      y1(j)=x(i);
      j=j+1;
  end
end
y2=y1/(max(abs(y1)));
y3=[y2,zeros(1,3120-length(y2))];
y=filter([1 -0.9],1,y3');  % high pass filter to boost the high frequency components

%%frame blocking

blocklen=240;%30ms block
overlap=80;
block(1,:)=y(1:240);
for i=1:18
    block(i+1,:)=y(i*160:(i*160+blocklen-1));
end
w=hamming(blocklen);
for i=1:19
    a=xcorr((block(i,:).*w'),12); % finding auto correlation from lag -12 to 12
    for j=1:12
        auto(j,:)=fliplr(a(j+1:j+12)); % forming autocorrelation matrix from lag 0 to 11
    end
    z=fliplr(a(1:12)); % forming a column matrix of autocorrelations for lags 1 to 12
    alpha=pinv(auto)*z';
    lpc(:,i)=alpha;
end
wavplay(x,Fs);
X1=reshape(lpc,1,228);
a1=sigmoid(Theta1*[1;X1']);
    h=sigmoid(Theta2*[1;a1]);
    m=max(h);
  p1=find(h==m);
  if(p1==10)
      P=0
  else
      P=p1
  end
end

Answer 1

[Greg Heath]

Consider this forced classification example

 rng(0)
 trueclassindices = [ randperm(10),randperm(10)]-1
 % [ 5 2 6 7 4 0 1 3 8 9 5 0 6 3 8 4 7 2 9 1 ]
 targets          = full( ind2vec(trueclassindices+1) )

targets are columns of the 10 dimensional unit matrix eye(10)

the inverse relation between the column vectors and the class indices is given by

 trueclassindices = vec2ind(targets)-1

The range of outputs will depend on the output transfer function:

 softmax: (0,1) , sum(outputs) = ones(1,10)
 logsig : (0,1) , sum(outputs) not constrained
 purelin:  neither range or sum is constrained

All are assumed to give "consistent" estimates of the class posterior probabilities, conditional on the input P(i|input). The assigned class is determined by the maximum output:

y =  net(x)
[ Pmax index ] = max(y)
assignedclassindex = index-1

However, for conditional classification, if Pmax is not large enough, a class assignment is not made

 if Pmax >= threshold
   assignedclassindex = index-1  % 0-9
 else
   assignedclassindex = 10
 end

 The threshold is determined from the training and validation data. 
 It may be prudent to add nonclass data to help determine the threshold.

 With forced classification

 outputs = [ y1 y2 y3...  y20]
 assignedclassindices = vec2ind(outputs)-1
 err = ( assignedclassindices  ~= trueclassindices)

% From this vector of zeros and ones, the error rates of each class can be determined. The total error rate is given by

 Nerr = sum(err); PctErr = 100*Nerr/N.

 However, with conditional classification, modifications would have to be made to include 
the nonclassification rate.

Hope this helps.

Thank you for formally accepting my answer

Greg

Answer 2

[Lakitha Hiran]

THANKS GREG FOR UR HELP ... I LL GO THROUGH THIS EXAMPLE AND DO NECESSARY CHANGES ... THANX A LOTT