Page 96 - Algorithm Collections for Digital Signal Processing Applications using MATLAB

84 Chapter 2

for k=1:1:length(x)
col= [col a(x(k),y(k))];
end
plot(col,zeros(1,length(col)),strcat(u(i),'*'))
hold on
end
pause(0.01)
end
figure
plot(s)
xlabel('Iteration')
ylabel('Change in the membership values
___________________________________________________________

5. MEAN AND VARIANCE NORMALIZATION

Suppose in the speech recognition system the two speech signals
corresponding to the same word are to be compared. Two signals are not
recorded exactly with the same volume (i.e.) the two signals are not with the
same energy. Thus before extracting features from the speech signals, there
is the need to normalize the speech signal in terms of mean and variance so
that two speech signals are made recorded with the same volume.
Mean and variance normalization of the signal is obtained as described
below. Consider the speech signal ‘A(t)’ with mean ‘m d’ and variance ‘v d’
and speech signal ‘B(t)’ with mean m and variance ‘v’. The requirement is to
normalize the speech signal B(t) so that the mean and variance are changed
to the desired mean ‘m d’ and desired variance ‘v d’ respectively. The
procedure for the mean and variance normalization is given below.

5.1 Algorithm

Step 1: Create the vector completely filled up desired mean ‘m d’. Number of
elements of the vector is equal to the number of samples in the
speech signal.
Step 2: Each and every element of the vector is added with +Δ or -Δ. If the
sample value in the original vector is greater than mean ‘m’, add +Δ
to the corresponding sample in the vector created. Otherwise the
value is added with ‘-Δ’. The value of the ‘Δ’ is computed using the
sample value, mean ‘m’, variance ‘v’ of the signal ‘A (t)’ and the
desired variance ‘v d’ .
th
Let Δ i be the value calculated for the i sample of the signal ‘A’
(original signal) and is computed as described below.