Application of the neural networks for text-to-phoneme mapping

In this paper we present the results on the use of neural networks for text-to-phoneme mapping. For this mapping, we have compared the performances of the Context Dependent Multilayer Perceptron network with the Recurrent Neural Network. The results (number of parameters vs neural network model vs phoneme accuracy) are given for American English. Also some guidelines for selecting the appropriate network structure together with some methods for improving the phoneme accuracy are presented.