Automatic speaker verification using the neural network and combined LPC parameters

Presents a new method of automatic speaker verification based on a neural network and using the combined LPC paramaters. In the study, a three layered feed-forward neural network is adapted and a variable step (VS) back-propagation algorithm for the automatic learning is used. Several sorts of LPC parameters and their combination are used as the input of the NN mentioned above. In the verification test of the experiment, three vowels /a/, /e/, /i/ uttered by each speaker are used as text-dependent speech, the population is 10 speakers (8 male and 2 female). The test shows the average correct rate of verification is 95 percent.<
>