Pattern matching apparatus

Claim 2

A pattern matching apparatus comprising: a first memory for storing an input patter A={a(1), a(2), . . . , a(i), . . . , a(I)}, which is expressed as a sequence of feature vectors; a second memory for storing a first reference pattern B 1 ={b 1 (1), b 1 (2), . . . , b 1 (j), . . . , b 1 (J 1 )} and a second reference pattern B 2 ={b 2 (1), b 2 (2), . . . , b 2 (j), . . . , b 2 (j 2 )}, which are expressed as sequences of feature vectors; a first calculator for calculating first and second quantities representing distances between said features a(i) and b 1 (j) and between said features a(i) and b 2 (j), respectively, at designated point (i, j); a second calculator for calculating first and second integrals g 1 (i,j) and g 2 (i,j) of said first and second quantities at the point (i, j) by a time normalization matching method utilizing dynamic programming on the basis of said first and second quantities at the point (i,j), said first integral being obtained in a positive time direction from (i,j)=(1,1) to (I,J 2 ) with respect to said first reference pattern, and said second integral being obtained in a negative time direction from (i,j)=(I 1 ,J 2 ) to (1,1) with respect to said second reference pattern; a third calculator for calculating an optimum integral at a designated i as a minimum or maximum value out of a set {g 1 (i, j), j=1, 2, . . . , J} of the integrals in the positive direction with respect to the integral g 1 (i,j) in a positive direction, and as a value of the integral g(i, 1) at a starting end (i.e. j=1) with respect to the integral g 2 (i,j) in the negative direction; a third memory for storing the optimum integral {g, (i), i=1, 2, . . . , I} in the positive direction and the optimum integral {g 2 (i), [=1, 2, . . . , I} in the negative direction; and a determination means for determining a whole quantity between said input pattern and a reference pattern which is obtained by connecting a front portion of said first reference pattern and a second reference pattern together as a minimum or maximum value out of the results obtained by calculating g*(i)=g 2 (i)+g 2 (i) for each i.

Claim 3

An apparatus according to claim 1, wherein said determination means develops a minimum or maximum value obtained by dividing the g*(i) by the weighted sum of the length of said input pattern and said first and second reference patterns.

Claim 4

An apparatus according to claim 2, wherein a minimum value of said normalized g*(i) is expressed by: ##EQU9##

Claim 5

An apparatus according to claim 1, wherein ##EQU10##

Claim 6

An apparatus according to claim 1, wherein ##EQU11##

Claim 7

A pattern matching apparatus for the recognition of speech, comprising: a first means for extracting a vowel portion from an input speech pattern expressed as a sequence of feature vectors; a second means for storing a predetermined portion of said input speech pattern and outputting an input pattern A={a(1), a(2), . . . , a(i), . . . , a(I)} as a syllable pattern V 1 CV 2 consisting of two extracted vowels V 1 , V 2 and a consonant C existing therebetween; a third means for storing syllable patterns consisting of combinations a predetermined number of first vowels, a predetermined number of consonants and second vowels with the order of these vowels and consonants maintained, the number of said second vowels being smaller than that of said first vowels, and outputting a first reference pattern B 1 {b 1 (1), b 1 (2), . . . , b 1 (j), . . . , b 1 (J 1 )} as a syllable pattern V 1 C n V 0 in response to said extracted vowel V 1 , designated consonant C n and second vowel V 0 ; a fourth means for storing syllable patterns consisting of combinations of a predetermined number of consonants and a predetermined number of third vowels with the order of these consonants and vowels maintained, and outputting a second reference pattern B 2 ={b 2 (1), b 2 (2), . . . , b 2 (j), . . . , b 2 (J 2 )} as a syllable pattern C n V 2 in response to said designated consonant C n and said extracted vowel V 2 ; a fifth means for calculating first and second quantities representing distances between said features a(i) and b 1 (j) and between said features a(i), respectively, at the designated point (i,j); a sixth means for calculating first and second integrals g 1 (i,j) of said first and second quantities at the point (i,j) by a time normalization matching method utilizing dynamic programming on the basis of said first and second quantities at the point (i,j), said first integral being obtained in a positive time direction from (i,j)=(1,1 to (I,J 1 ) with respect to said first reference pattern, and said second integral being obtained in a negative time direction from (i,j)=(I,J 2 ) to (1,1) with respect to said second reference pattern; a seventh means for determining an optimum integration amount for a designated i as a minimum or maximum value out of a set {g(i,j), j=1, 2, . . . , J} of the integration amount in the positive direction with respect to the integration amount g 1 (i,j) in the positive direction, and as a value of the integration amount g(i,j) at a terminal end point (i.e. j=1) with respect to the integration amount g 2 (i,j) in the negative direction; an eighth means for storing the optimum integration amount {g 1 (i), i=1, 2, . . . , I} in the positive direction and optimum integration amount {g 2 (i), i=1, 2, . . . , I} in the negative direction; a ninth means for determining g*(i)=g 1 (i)+g 2 (i) for each i with all of the consonants designated, to obtain a minimum or maximum value out of the results of the determination; and a tenth means for determining a designated consonant giving the minimum or maximum value of g* as a consonant included in said input syllable pattern.

Claim 8

A pattern matching apparatus for the recognition of speech, comprising: a first means for extracting a vowel portion from an input speech pattern expressed as a sequence of feature vectors; a second means for storing a predetermined portion of said input speech pattern and outputting an input pattern A={a(1), a(2), . . . , a(i), . . . , a(I)} as a syllable pattern V 1 CV 2 consisting of two extracted vowels V 1 , V 2 and a consonant C existing therebetween; a third means for storing syllable patterns consisting of combinations a predetermined number of first vowels, a predetermined number of consonants and second vowels with the order of these vowels and consonants maintained, the number of said second vowels being smaller than that of said first vowels, and outputting a first reference pattern B 1 {b 1 (1), b 1 (2), . . . , b 1 (j), . . . , b 1 (J 1 )} as a syllable pattern V 1 C n V 0 in response to said extracted vowel V 1 , designated consonant C n and second vowel V 0 ; a fourth means for storing syllable patterns consisting of combinations of a predetermined number of consonants and a predetermined number of third vowels with the order of these consonants and vowels maintained, and outputting a second reference pattern B 2 ={b 2 (1), b 2 (2), . . . , b 2 (j), . . . , b 2 (J 2 )} as a syllable pattern C n V 2 in response to said designated consonant C n and said extracted vowel V 2 ; a fifth means for calculating first and second quantities representing distances between said features a(i) and b 1 (j) and between said features a(i), respectively, at the designated point (i,j); a sixth means for calculating first and second integrals g 1 (i,j) of said first and second quantities at the point (i,j) by a time normalization matching method utilizing dynamic programming on the basis of said first and second quantities at the point (i,j), said first integral being obtained in a positive time direction from (i,j)=(1,1 to (I,J 1 ) with respect to said first reference pattern, and said second integral being obtained in a negative time direction from (i,j)=(I,J 2 ) to (1,1) with respect to said second reference pattern; a seventh means for determining an optimum integration amount for a designated i as a minimum or maximum value out of a set {g(i,j), j=1, 2, . . . , J} of the integration amount in the positive direction with respect to the integration amount g 1 (i,j) in the positive direction, and as a value of the integration amount g(i,j) at a terminal end point (i.e. j=1) with respect to the integration amount g 2 (i,j) in the negative direction; an eighth means for storing the optimum integration amount {g 1 (i), i=1, 2, . . . , I} in the positive direction and optimum integration amount {g 2 (i), i=1, 2, . . . , I} in the negative direction; a ninth means for determining g*(i)=g 1 (i)+g 2 (i) for each i with all of the consonants designated, to obtain a minimum or maximum value out of the results of the determination; and a tenth means for determining a designated consonant giving the minimum or maximum value of g* as a consonant included in said input syllable pattern.

Claim 9

An apparatus according to claim 6, wherein said second vowel is one predetermined vowel.

Claim 10

An apparatus according to claim 6, wherein semivowels are included as said first and second vowels.

Claim 11

An apparatus according to claim 6, wherein said input pattern is a syllable pattern having a boundary in substantially the center of an extracted vowel.