A study on vowel behaviour and its description by a statistical model

A complete statistical model for italian vowels is presented. The classification method contains the adaptive procedure which makes possible the evolution of the regions characteristic of each vowel in the F1-F2 plane and the effectiveness of which has been verified in a previous work [1]. In this paper, each part of the algorithm is described in its theoretical detail and justified on the basis of an extended experimentation. Results are given proving the accuracy of the classifier for vowel segments independent of the context and prosody and dependent upon the context and prosody.