A method of spoken word recognition has been proposed using utterance normalization, multiple templates, and path-limited DP matching to cope with speaker variability as well as to reduce the amount of computation. This paper shows the overall configuration of the system and the design of the template-matching part as a cyclic systolic array that can be implemented on an available gate array. Analysis of system operation by a circuit simulator showed that the system is capable of real time recognition of words from a 3,000-word vocabulary, assuming the use of eight templates per word.
Keywords: Spoken Word Recognizer, Utterance Normalization, Multiple Templates, Path-Limited DP Matching, Cyclic Systolic Array