Nonlinear Adaptive Speech Prediction using a Pipelined Recurrent Fuzzy Network

In this paper, a pipelined TSK-type recurrent fuzzy network (PTRFN) is proposed for nonlinear adaptive signal prediction. The PTRFN model consists of a number of modules interconnected in a cascaded form. The participating modules are implemented through recurrent fuzzy neural networks with internal dynamics. The structure of the modules is evolved sequentially from input-output data. The parameter learning task is accomplished using a gradient descent algorithm and the extended least squares method. The suggested predictor exhibits a series of attractive attributes, including effective spatial representation of the temporal patterns, enhanced memorizing capabilities, and low computational complexity. The nonlinear subsection of the predictor (PTRFN), followed by a linear subsection (a tapped delay-line filter) is tested on the adaptive speech prediction problem. Simulation results demonstrate that considerably better performance is obtained compared with other existing recurrent networks