System for recognizing words continuously spoken according to a format

A system for recognizing an input pattern feature vector time sequence A(0, I) consisting of a certain number I of first through I-th input pattern feature vectors representative of an input word sequence consisting of first through k-th (K being representative of a predetermined integer) input word units selected from first through K-th word sets, respectively, as one of a plurality of reference word sequences, each word set consisting of at least one reference word unit selected from first through N-th (N being representative of a predetermined number) reference word units B.sup.n +s (n being representative of each of 1 through N), each reference word sequence consisting of first through K-th reference word units selected from the respective word sets, said system comprising:

• input pattern buffer memory means for memorizing at least a preselected plurality of successive input pattern feature vectors at one time;
• reference pattern memory means for memorizing said reference word units as a plurality of reference pattern feature vector time sequences, respectively, and for producing said reference word units successively in a predetermined order;
• similarity measure calculating means connected to said input pattern buffer memory means and said reference pattern memory means for calculating a plurality of elementary similarity measures S(p, q, n)'s between a plurality of elementary partial patterns A(p, q)'s and the successively produced reference word units, each partial pattern being segmented from said input pattern feature vector time sequence at a pair of smaller-numbered and greater-numbered segmentation points p and q so as to consist of (p+1)-th through q-th input pattern feature vectors;
• format memory means for memorizing a format defined by said word sets;
• word set specifying means connected to said format memory means for successively specifying said word sets in a prescribed order to make said format memory means produce the reference word units of the specified word sets;
• and
• matching means connected to said similarity measure calculating means, said format memory means, and said word set specifying means for carrying out pattern matching between a plurality of specific partial pattern concatenations and said plurality of reference word sequences to decide said one reference word sequence by finding a maximum of sums of individual similarity measures S(p(k-1), p(k), n(k))'s (k being representative of each 1 through K) with each individual similarity measure selected, while a k-th word set is specified, from said plurality of elementary similarity measures with each reference word unit B.sup.n(k) selected from said k-th word set and with a k-th specific partial pattern A(p(k-1), p(k)) in each specific partial pattern concatenation selected from said plurality of elementary partial patterns.