Improvement and discussion on pronunciation method of DIVA model based on auditory perception space

Abstract

Based on DIVA(Directions Into of Articulators) model, the paper attempts to construct a neural network model which can better simulate speech perception, speech acquisition and speech production. The model classifies speech phonemes from formant frequencies by unsupervised learning, and through the modifications of DIVA model speech mapping and self-organizing mapping algorithm, it simulates the natural process of human speech acquisition in auditory. What's more, by the comparison with achievements of psychological phonetics experiments, the simulation result shows that the model we build can better identify the difference between prototype and non-prototype and fully form auditory perception for vowels, so it better shows the natural process of the ability of human speech acquisition.