Recognition of Assamese Spoken Words using a Hybrid Neural Framework and Clustering Aided Apriori Knowledge

摘要

In this paper, an Artificial Neural Network (ANN) based model is proposed for recognition of discrete Assamese speech using a Self Organizing Map (SOM) based phoneme count determination technique. The phoneme count determination technique takes some initial decision about the possible number of phonemes in the word to be recognized and accordingly the word is presented to some N-phoneme recognition algorithm. In this paper recognition algorithm is designed to recognize three phoneme consonant-vowel-consonant (CVC) type Assamese words. The word recognizer is consisted of another SOM block to provide phoneme boundaries and Probabilistic Neural Network (PNN) and Learning Vector Quantization (LVQ) to identify the SOM segmented phonemes. The recognition of constituent phonemes in turn represents the discrimination between incoming words with a minimum success rate of 90%.