5th International Conference on Spoken Language Processing

Sydney, Australia
November 30 - December 4, 1998

Periodicity Emphasis of Voice Wave using Nonlinear IIR Digital Filters and Its Applications

Hiroyuki Kamata, Akira Kaneko, Yoshihisa Ishida

Meiji University, Japan

We propose a new method for emphasizing the periodicity of voice wave using chaotic neurons, and propose a practical method to detect the fundamental frequency of human voice. The chaotic neuron is a kind of nonlinear recursive mapping proposed in the field of nonlinear theory and is usually used to generate the chaotic signal. Besides, when the chaotic neuron is considered in the theory of linear signal processing, we can interpret that the chaotic neuron is a positive feedback IIR digital filter of first order, therefore, it gives a spectrum slope to the target spectrum of input speech signal. In this study, we try to tune up the chaotic neuron to amplify the low frequency components to emphasize the component of fundamental frequency. As the result, spectrum peaks based on the formants are canceled, the spectrum peak corresponded to the fundamental frequency of voiced speech can be detected easily. In addition, a nonlinear function that has a dead band is included in the feedback loop of the chaotic neuron. As its effect, noise components of unvoiced speech are not amplified.

Full Paper

Bibliographic reference.  Kamata, Hiroyuki / Kaneko, Akira / Ishida, Yoshihisa (1998): "Periodicity emphasis of voice wave using nonlinear IIR digital filters and its applications", In ICSLP-1998, paper 1016.