The
previous
groups
selected for the INN was
the group no. 1 (2, 4, 8, 9,
10), and group No. 2 (0, 1,
3, 5, 6, 7). They were of less
recognition rate (Table 3),
and so some new pairs are
studied as given in Table 4.
A BP classifier network [5,
6] is trained for classifying
the input in the chosen
pair group with varying
hidden nodes the result
of that test is summarized
in Fig. 5. Results for the
six
unknown
speakers appropriate
minimum selected
features
are
are summarized in Fig. 6. number
of
features non-parametric
time
domain features as ZCR,
maximum, and minimum,
but they were calculated
for each of the 15
segments of the words.
Whatever, new features enough for the speech
were selected to find the recognition. The new
Research
35

---