Method for protecting biometric templates, and a system and method for verifying a speaker's identity
View Patent ↗A method for protecting a biometric template, comprising the steps of: retrieving an original vector (V) representing said biometric template, said vector comprising a plurality of original elements (v 1 , v 2 , . . . v i , . . . , v n ); mapping at least some elements from said original vector to a protected vector (P) comprising a plurality of protected elements (p 1 , p 2 , . . . p i , . . . , p n−m+1 ), the mapping being based on multivariate polynomials defined by m user-specific coefficients (C) and exponents (E).
1. A method for protecting a biometric template, said biometric template being a set of salient features representing a particular biometric characteristic of an individual, comprising the steps of:
retrieving an original vector representing said biometric template, said original vector comprising n original elements, each original element being a real number, wherein said step of retrieving comprises recording with a microphone, converting into digital signal, pre-processing, extracting features, and modelling;
mapping at least some original elements from said original vector to a protected vector comprising a plurality of n-m+1 protected elements, each protected element being a real-number;
said mapping being based on multivariate polynomials defined by a number m of user-specific coefficients and exponents, said multivariate polynomials being of the form:
p i =c 1 v i e1 +c 2 v i+1 e2 + . . . +c m v i+m-1 em
where m<n is an integer, 1 i≤n-m+1, [c 1 , c 2 , . . . , c m ] are the user-specific coefficients and [e 1 , e 2 , . . . , e m ] are the user-specific exponents.
2. The method of claim 1 , wherein said original vector is an i-vector.
3. The method of claim 1 , wherein said original vector represents a voice biometric template.
4. The method of claim 1 , said step of mapping comprising mapping overlapping sets of consecutive original elements to single protected elements via multivariate polynomials defined by a set comprising a number of user-specific coefficients and exponents.
5. The method of claim 1 , wherein the dimensionality of said protected vector is lower than the dimensionality of said original vector.
6. The method of claim 1 , said biometric template being a test biometric template, the method further comprising a step of matching said protected vector with a protected reference vector corresponding to an enrolment biometric template.
7. The method of claim 6 , said step of matching comprising computing a cosine similarity between said protected vector (P Q ) and said protected reference vector.
8. A method for verifying or establishing a speaker's identity, comprising:
retrieving a speaker's test speech sample;
determining from said speaker's test speech sample an original vector representing a biometric template of said speech sample, said original vector comprising n original elements each original element being a real number;
mapping at least some original elements from said original vector to a protected test vector comprising a plurality n-m+1 of protected elements, each protected element being a real-number, the mapping being based on multivariate polynomials defined by a number m of user-specific coefficients and exponents, said multivariate polynomials being of the form:
p i =c 1 v i e1 +c 2 v i+1 e2 + . . . +c m v i+m-1 em
where m<n is an integer, 1≤i<n-m+1, [c 1 , c 2 , . . . , c m ] are the user-specific coefficients and [e 1 , e 2 , . . . , e m ] are the user-specific exponents,
matching said protected test vector with at least one protected reference vector representing a voice biometric template.
9. The method of claim 8 , said step of retrieving a test speech sample comprising recording with a microphone and converting to a digital signal.
10. The method of claim 9 , said step of determining from said speaker's test speech sample an original vector representing biometric features of said speech sample comprising pre-processing said digital signal and extracting features from the pre-processed digital signal.
11. A system for verifying or establishing a speaker's identity, comprising:
a module for retrieving a speaker's test speech sample;
a module arranged for determining from said speaker's test speech sample an original test vector representing a biometric template of said test speech sample, said original test vector comprising n original elements, each original element being a real number;
a module arranged for mapping at least some of the original elements from said original vector to a protected test vector comprising a plurality of m-m+1 protected elements, each protected element being a real-number, the mapping being based on multivariate polynomials defined by a number m of user-specific coefficients and exponents, said multivariate polynomials being of the form:
p i =c 1 v i e1 +c 2 v i+1 e2 + . . . +c m v i+m-1 em
where m<n is an integer, 1≤i<n-m+1, [c 1 , c 2 , . . . , c m ] are the user-specific coefficients and [e 1 , e 2 , . . . , e m ] are the user-specific exponents;
a module arranged for matching said protected test vector with at least one protected reference vector representing a voice biometric template.
12. The system of claim 11 , said module for retrieving a speaker's test speech sample comprising a microphone.
13. The system of claim 11 , said module arranged for determining from said speaker's test speech sample an original vector comprising a processor and a memory comprising a software code portion.