A recurrent fuzzy neural network for adaptive speech prediction
Dimitris G. Stavrakoudis, John B. Theocharis · 2007
An enhanced memory TSK-type fuzzy neural network (EM-TRFN) is proposed in this paper, suitable for nonlinear adaptive speech prediction. The feedback links of the network are realized through finite impulse response (FIR) synapses, increasing the depth of the time-series history the network processes. The EM-TRFN is evolved in an on-line manner, with concurrent structure and parameter learning. Simulations on a speech signal prediction problem illustrate the effectiveness of the proposed network, provided by its enhanced temporal capabilities, in grasping the complex dynamic of the speech signal.