Claims Analysis
Total Claims: 9
Claim 1
Independent Claim
136 words
A speech recognition method, comprising the steps of: inputting a speech signal; analyzing said input speech signal and outputting feature parameters for each of a pluraity of successive analysis frame periods of the speech represented by said speech signal; generating information of a power difference indicating whether the speech power of said speech is increasing or decreasing for each analysis frame; calculating the similarity between frames of said speech and a standard pattern stored beforehand by using said feature parameters and further feature parameters obtained from said standard pattern; correcting said calculated similarity based on said power difference information and further power difference information of said standard pattern, which is stored beforehand; and matching said speech and said standard pattern by using said corrected calculated similarity and recognizing said speech from the result of said matching.
Claim 2
Dependent Claim
136 words
A speech recognition method, comprising the steps of: inputting a speech signal; analyzing said input speech signal and outputting feature parameters for each of a pluraity of successive analysis frame periods of the speech represented by said speech signal; generating information of a power difference indicating whether the speech power of said speech is increasing or decreasing for each analysis frame; calculating the similarity between frames of said speech and a standard pattern stored beforehand by using said feature parameters and further feature parameters obtained from said standard pattern; correcting said calculated similarity based on said power difference information and further power difference information of said standard pattern, which is stored beforehand; and matching said speech and said standard pattern by using said corrected calculated similarity and recognizing said speech from the result of said matching.
Claim 3
Dependent Claim
44 words
A speech recognition method according to claim 1 wherein said correction step comprises selecting one of a number of predetermined weights based on the result of said power difference information obtained from said generating step and said calculated similarity obtained from said caluculating step.
Claim 4
Dependent Claim
32 words
A speech recoginition method according to claim 2 wherein said correcting step comprises multiplying said selected weight by said calculated similarity obtained in said calculating step to determine the corrected calculated similarity.
Claim 5
Dependent Claim
49 words
A speech recognition method according to claim 1 wherein said feature parameters of said speech are determined utilizing autocorrelation coeffients thereof for inter-frames and sasid calculated simlarity is determined by calculating a normalized residual power from a function having variables of the auto-correlation coefficients and the standard pattern data.
Claim 6
Dependent Claim
36 words
A speech recognition method according to claim 1 wherein an ouput value from a band-pass filter is used as a feature parameter of said input speech and a distance is adopted as said degree of similarity.
Claim 7
Dependent Claim
69 words
A speech recognition method according to claim 1, wherein said calculation of the similarity between said speech and a standard pattern is performed in synchronism with calculation of the similarity between said speech and at least one partial standard pattern specified beforehand for said standard pattern with respect to a predetermined time base and said speech is determined on the basis of the total results obtained from said calculations.
Claim 8
Dependent Claim
26 words
A speech recognition method according to claim 6 wherein a nonlinear time warping matching based on a dynamic programming method is utilized as said similarity calculation.
Claim 9
Dependent Claim
42 words
A speech recognition method according to claim 6, comprising a step of determining said speech on the basis of the sums of the similarity calculations with respect to respective standard patterns and the similarity calculations with respect to respective partial standard patterns.