Accurate Parameter Estimation for an Articulatory Speech Synthesizer with an Improved Neural Network Mapping

摘要

Neural network (NN) applications have recently been employed toextract the parameters of an articulatory speech synthesizer from agiven speech signal. Results from these attempts showed that a singleNN is insufficient to cover all of the possible configurationsuniquely. Moreover, apart from their computational advantages, NNmapping is so far not superior to the other mapping techniques [1]. Thus there is a clear need to improve NN solution to the inverseproblem.Results from our earlier experiments with an articulatory speechsynthesizer have shown that the statistical characteristic of thearticulatory target pattern vectors can be exploited for animprovement in the estimation performance of a Multi-Layer Perceptron(MLP) NN [2]. In this paper, the effect of the modification to thedistribution characteristic of the acoustic input pattern vectors willbe investigated. The theoretical background for the effect of the inputdistribution characteristics on neural learning has been detailed elsewhere[3]. Empirical results for a more correct estimation ofarticulatory speech synthesizer parameters through exploiting thebehavior of the Back Propagation (BP) algorithm are focused on here.