PROPOSAL OF A KNOWLEDGE BASED ISOLATED WORD RECOGNITION.

The authors describe a knowledge-based isolated Japanese word recognition algorithm. The program is written in Prolog/KR and has two basic inference processes: a bottom-up search and a top-down search. In the bottom-up process, segmentation and vowel decision are performed and some target word patterns are generated. The top-down process includes a consonant decision using a score for each candidate word calculated on the basis of fuzzy set theory. In the vowel inference step, a template matching is applied. In the segmentation step, heuristic rules based on the spectrum transition and the wave form are used. In the consonant inference step, each rule has a hierarchical structure and the consonant is defined automatically in the form of a multivalued threshold function from learning data.