Segmentation and Classification of Vowel Phonemes of Assamese Speech Using a Hybrid Neural Framework

In spoken word recognition, one of the crucial point is to identify the vowel phonemes. Vowel phonemes are used to combine two or more consonant phonemes in most of the words spoken, and the meaning of the words changes with the change of vowels. Therefore, in order to recognize a word, identification of vowel phoneme is as important as the identification of constituent consonant phonemes. This paper describes an Artificial Neural Network (ANN) based algorithm developed for segmentation of vowel phonemes of Assamese language from some words containing those vowels. Self Organizing Map (SOM) trained with various number of iterations are used to segment the word into its constituent phonemes. Later, Probabilistic Neural Network (PNN) trained with clean vowel phonemes, are used to recognize the specific vowel segments from the six different SOM segmented phonemes. One of the important aspects of the proposed algorithm is that it proves the validation of the recognized vowel by checking its first formant frequency. The first formant frequency of all the Assamese vowels are predetermined by estimating pole or formant location from the linear prediction (LP) model of vocal tract. The proposed algorithm shows higher recognition performance in comparison to the conventional Discrete Wavelet Transform (DWT) based segmentation.