سال انتشار: ۱۳۸۲

محل انتشار: یازدهمین کنفرانس مهندسی پزشکی ایران

تعداد صفحات: ۶

نویسنده(ها):

R. Fazel-Reazi – Department of Electrical Engineering, University of Manitoba, Canada
Sh. Oveisgharan – Department of Electrical Engineering, Sharif University of Technology, Iran

چکیده:

A chaotic and a fractal measures were calculated for Persian speech signal and their performances in speech classification were evaluated. The first measure was correlation dimension of each frame in speech signal which is based on its chaotic characteristics. The second measure was fractal dimension computed by fitting Hosking’s ARMA filtered FdGn model [1] to speech signal and computing its Hurst parameter by Tewfik’s iterative Maximum Likelihood approach [2]. Experimental results showed that, in Persian speech, better classification were obtained by Hosking’s model because its ability to characterize short term dependencies of speech signal which is interpretable by ARMA model.