High speed recognition of a string of words connected according to a regular grammar by DP matching

Claim 1

A method of recognizing, as an optimum one of word concatenations, an input string of words represented by an input sequence of input pattern feature vectors, said input string of words being selected from a word set of first through N-th words and continuously uttered in compliance with a regular grammar, where N represents an integer greater than one, each concatenation being a string of words selected from said word set and concatenated in compliance with said grammar, said grammar specifying a set of transition rules of said first through said N-th words and a set of final states of said concatenations, an n-th word optionally selected from said first through said N-th words of said word set appearing in a plural number K of said transition rules as said n-th word of first through K-th occurrence, each of the n-th word of said first through said K-th occurrence defining a combination of said n-th word and a state pair consisting of a start and an end state of said n-th word, said method comprising the steps of: memorizing reference pattern feature vectors representative of said first through said N-th words, the reference pattern feature vectors representative of said n-th word being feature vectors representative of said n-th word being consecutively numbered as first through J-th feature vectors according to utterance of said n-th word, where J represents an integer greater than one; generating a basic timing signal successively specifying first through I-th periods assigned to the respective input pattern feature vectors, a word specifying signal specifying, while an i-th period is specified as each of said first through said I-th periods, said first through said N-th words, and a state specifying signal specifying, while said i-th period is specified, the state pairs of said first through said N-th words, where I represents an integer greater than one; selecting in response to the input sequence and the base timing, the word specifying, and the state specifying signals, while said n-th word is specified in said i-th period, one of the n-th word of said first through said K-th occurrence as a selected n-th word; calculating in response to the input sequence and the base timing, the word specifying, and the state specifying signals, while said n-th word is specified in said i-th period, a plurality of distances between the input pattern feature vector assigned to said i-th period and said first through said J-th feature vectors only for said selected n-th word; calculating in response to the base timing, the word specifying, and the state specifying signals, while said n-th word and each state pair of said n-th word are specified in said i-th period, a recurrence value g(n, i, j) only for said selected n-th word and the state pair for said selected n-th word under a boundary condition iteratively in correspondence to said first through said J-th feature vectors by using the distance calculated for said selected n-th word before said i-th period, where n represents said n-th word, i represents said i-th period, and j represents a j-th feature vector representative of each of the first through the J-th feature vectors, said last-mentioned calculating step estimating the recurrence value for said selected n-th word by using the recurrence value for said selected n-th word and the boundary conditions for said selected and said nonselected n-th word, said nonselected n-th word being a remaining one of the n-th word of said first through said K-th occurrence that is not selected by said selecting step, said last-mentioned calculating step finding, while said i-th period is specified, a minimum T(q, i) of the recurrence values obtained for the words having state pairs including said end state q of said each state pair, said last-mentioned calculating step deciding that particular word and that particular start state of said particular word for which said minimum is found, said boundary conditions being given by T(p, 1), where p represents the start state of said each state pair; selecting in response to the basic timing, the word specifying, and the state specifying signals, while said i-th period is specified, a particular period from said first through said (i-1)-th periods with reference to said particular start state and said particular word; and deciding the optimum concatenation by referring, after lapse of said I-th period, to the minima found in said first through said I-th periods, respectively, and to those particular words, those particular start states, and those particular start periods which are decided in said first through said I-th periods; said first-mentioned selecting step being for selecting, as said selected n-th word, said one of the n-th word of said first through said k-th occurrence that has a minimum value of the boundary conditions T(p, i-1)'s.

Claim 2

A method of recognizing, as an optimum one of word concatenations, an input string of words represented by an input sequence of input pattern feature vectors, said input string of words being selected from a word set of first through N-th words and continuously uttered in compliance with a regular grammar, where N represents an integer greater than one, each concatenation being a string of words selected from said word set and concatenated in compliance with said grammar, said grammar specifying a set of transition rules of said first through said N-th words and a set of final states of said concatenations, an n-th word optionally selected from said first through said N-th words of said word set appearing in a plural number K of said transition rules as said n-th word of first through K-th occurrence, each of the n-th word of said first through said K-th occurrence defining a combination of said n-th word and a state pair consisting of a start and an end state of said n-th word, said method comprising the steps of: memorizing reference pattern feature vectors representative of said first through said N-th words, the reference pattern feature vectors representative of said n-th word being feature vectors representative of said n-th word being consecutively numbered as first through J-th feature vectors according to utterance of said n-th word, where J represents an integer greater than one; generating a basic timing signal successively specifying first through I-th periods assigned to the respective input pattern feature vectors, a word specifying signal specifying, while an i-th period is specified as each of said first through said I-th periods, said first through said N-th words, and a state specifying signal specifying, while said i-th period is specified, the state pairs of said first through said N-th words, where I represents an integer greater than one; selecting in response to the input sequence and the base timing, the word specifying, and the state specifying signals, while said n-th word is specified in said i-th period, one of the n-th word of said first through said K-th occurrence as a selected n-th word; calculating in response to the input sequence and the base timing, the word specifying, and the state specifying signals, while said n-th word is specified in said i-th period, a plurality of distances between the input pattern feature vector assigned to said i-th period and said first through said J-th feature vectors only for said selected n-th word; calculating in response to the base timing, the word specifying, and the state specifying signals, while said n-th word and each state pair of said n-th word are specified in said i-th period, a recurrence value g(n, i, j) only for said selected n-th word and the state pair for said selected n-th word under a boundary condition iteratively in correspondence to said first through said J-th feature vectors by using the distance calculated for said selected n-th word before said i-th period, where n represents said n-th word, i represents said i-th period, and j represents a j-th feature vector representative of each of the first through the J-th feature vectors, said last-mentioned calculating step estimating the recurrence value for said selected n-th word by using the recurrence value for said selected n-th word and the boundary conditions for said selected and said nonselected n-th word, said nonselected n-th word being a remaining one of the n-th word of said first through said K-th occurrence that is not selected by said selecting step, said last-mentioned calculating step finding, while said i-th period is specified, a minimum T(q, i) of the recurrence values obtained for the words having state pairs including said end state q of said each state pair, said last-mentioned calculating step deciding that particular word and that particular start state of said particular word for which said minimum is found, said boundary conditions being given by T(p, 1), where p represents the start state of said each state pair; selecting in response to the basic timing, the word specifying, and the state specifying signals, while said i-th period is specified, a particular period from said first through said (i-1)-th periods with reference to said particular start state and said particular word; and deciding the optimum concatenation by referring, after lapse of said I-th period, to the minima found in said first through said I-th periods, respectively, and to those particular words, those particular start states, and those particular start periods which are decided in said first through said I-th periods; said first-mentioned selecting step being for selecting, as said selected n-th word, said one of the n-th word of said first through said k-th occurrence that has a minimum value of the boundary conditions T(p, i-1)'s.