Discrete-word recognition utilizing a word dictionary and phonological rules

A discrete-word recognition system utilizing a word dictionary and phonological rules is described. In this system, nine distinctive features are extracted from a discrete-word input. Segmentation is performed using these features. Segmentation errors are corrected by applying a phoneme connecting rule. The input word is transformed into an input feature matrix. The comparison of this matrix with the standard derived from the dictionary is performed in the feature (matrix) space. Another method of segmentation is also described in which segmentation is performed using a duration dictionary. The effectiveness of utilizing a word dictionary and phonological rules in automatic discrete-word recognition is discussed.