Two
of
the
female second speaker has three For generalization, two
speakers were talking fast recognized words (27.3 speakers
(male
and
(Table 2), the network %). He uttered the words female) are trained for
recognized
only
one five and ten fast where RR 44 words (22 words/ each
word for each of them for the recognized three speaker) based on three
(the word representing words for the third speaker layers network, with 180
numbers 4 and 5). For the was (27.3 %). He uttered input nodes (12 LPC /15
third female speaker three the words two, seven, segments) and one output
words were recognized and nine unclearly, and node through variable
(27.3 %). She stressed the word one fast [3, 4]. hidden nodes (50, 75, 90).
Fig. 1 shows the results of
these experiments, but Fig.
2 illustrates the results of
the six unknown speakers.
The tests are repeated
with 362 input data and
the results are drawn in Fig.
3 where 75 hidden nodes
give the best rate (62.5 %).
Fig. 4 gives the deduced
curves of tests for the
six unknown speakers.
in uttering the words six
(settah, at s, and t), seven
(sabaah, at s), and nine
(tesaah, at s), she uttered
the word representing
one fast, and the word
representing ten unclearly.
For male speakers, the
network recognized four
words of the first speaker
(36.4
%),
he
uttered
the word ten unclearly,
and stressed the word
zero (sefr, at the s). The
Research
34

---