Method of comparing utterances for security control

• US 6,519,565 B1
• Filed: 11/10/2000
• Issued: 02/11/2003
• Est. Priority Date: 11/10/1998
• Status: Expired due to Term

• Current Assignee: Voice Security Systems Incorporated
• Original Assignee: Voice Security Systems Incorporated
• Inventors: Clements, Bradley Alan, Adcock, Sherrie, Schwartz, Gerald S.
• Primary Examiner(s): MCFADDEN, SUSAN IRIS
• Application Number:
  US09/710,590
• Time in Patent Office: 823 Days
• Field of Search: 704/251, 704/254, 704/273, 704/274, 704/275
• US Class Current: 704/273
• CPC Class Codes: G10L 17/00 Speaker identification or verification techniques
• CPC Class Codes: G10L 17/04 Training, enrolment or model building
• CPC Class Codes: G10L 17/06 Decision making techniques; Pattern matching strategies

Abstract

A security method compares a present verbal utterance with a previously recorded verbal utterance by comparing time-frequency domain representations of the utterances, with multiple repeat utterances forming a basis for determining a variation in repetitious performance by an individual, and similar differences between enrollment and challenge utterances forming a basis for a similar analysis of variance between enrollment and challenge utterances. In one embodiment a set of enrollment data is searched by each challenge until either a match is made, indicating an action, possibly dependent upon the specific match, or no match is made indicating an abort. In one application an individual is accepted or rejected as an imposter, in another applicaton, a selected action is accepted as corresponding to a verbal command.

Claims

1. A method of comparing a challenge verbal utterance with several enrollment verbal utterances from the same individual to determine a security action, the method comprising the steps of:

A) preparing an enrollment data comprising the steps of;

a) creating a Fast Fourier transform of an electrical signal representation of a verbal enrollment utterance to produce an enrollment frequency domain representation thereof;

b) taking M enrollment samples of the enrollment frequency domain representation for each of N frequency channels to form an M by N enrollment matrix E;

c) determining if at least X of the M enrollment samples have a selected minimum energy content in at least Y of the N frequency channels;

d) repeating steps (a) through (c) if the determination at step (c) is negative, and aborting after a selected number of negative repeats of steps (a) through (c);

e) repeating steps (a) through (c) if the determination at step (c) is positive, until a selected number of said sample enrollment matrices Ei are formed as E1, E2, E3, . . . ;

f) calculating the sums of the differences EDi between each Ei, as ED1=(E1−

E2)+(E1−

E3)+ . . . , ED2=(E2−

E1)+(E2−

E3)+ . . . ;

g) comparing the enrollment samples pairwise by moving each sample forward and backward in time to obtain a best match;

h) calculating an enrollment span, ES, equal to the difference between the largest and the smallest value of EDi; and

i) accepting the enrollment data if;

i. ES<

R1, and

ii. EDi<

R2 exists, where R1 and R2 are selected enrollment criterion,

B) challenging the enrollment data comprising the steps of;

a) creating a Fast Fourier transform of an electrical signal representation of a verbal challenge utterance to produce a challenge frequency domain representation thereof;

b) taking M challenge samples of the challenge frequency domain representation for each of N frequency channels to form an M by N challenge matrix C;

c) determining if at least X of the M challenge samples have a selected minimum energy content in at least Y of the N frequency channels;

d) repeating steps (a) through (c) if the determination at step (c) is negative, and aborting after a selected number of negative repeats of steps (a) through (c);

e) calculating the sums of the difference CDi between C and each Ei, as CD1=(C−

E1)+(C−

E2), CD2=(C−

E2)+(C−

E3), CD3=(C−

E3)+(C−

E4), . . . ;

f) comparing each one of the challenge samples pairwise by moving a the samples forward and backward in time to obtain a best match;

g) calculating a challenge span, CS, equal to the difference between the largest and the smallest value of CDi; and

h) accepting the challenge data only if;

i. CS<

V1, and

ii. CDi<

V2, exists, where V1 and V2 are selected verification criterion,

C) enabling the security action if the challenge verbal utterance is accepted.

2. The method of claim 1 further comprising the plural repeating of steps (A) (a) through (A) (i) for selected individuals so as to create a plurality of individual enrollment data sets.

3. The method of claim 2 further comprising the step of pairwise comparing the plurality of individual enrollment data sets and rejecting each new one of said enrollment data sets if corresponding enrollment spans do not differ by a selected amount.

4. The method of claim 2 further comprising the step of pairwise comparing the plurality of individual enrollment data sets and rejecting each new one of said enrollment data sets if corresponding enrollment spans differ by more than a selected amount.

5. The method of claim 1 wherein each of the data sets includes data defining a specific security action wherein upon after a successful verification, a further step is taken to complete said specific security action, said action including the generating of an electrical signal.

6. The method of claim 1 further comprising the plural repeating of steps (A) (a) through (A) (i) for selected actions so as to create a plurality of action enrollment data sets.