Feature selection and pattern recognition for language structure classification

One paradigm of the language identification problem is to first classify speech segments into a symbol string that adequately represents the language structure and then classify the symbol string for language identification. A method for automatically segmenting speech into several major structural (symbol) groups is given. Phonetically based structural groups are defined; LPC1 and LPC5 are selected as features to represent the speech; and a Bayes classifier is designed to automatically classify speech into these symbol groups. An example of speech sorted into these structural groups and the corresponding classifier design are shown