RBF networks vs. multilayer perceptrons for sequence recognition

In this paper we consider several learning procedures for radial basis function (RBF) networks applied to the problem of speech recognition. The dynamic nature of speech is considered by adding delayed connection and integration units to the network. Our study shows that the supervised learning of the centroids of the basis functions gives appreciable results at a significantly small cost. The results thus obtained are compared with the generalization performance of multilayer perceptrons. The possibility to include recurrent connections into RBF networks is also investigated.<>