Apparatus and methods for speaker verification/identification/classification employing non-acoustic and/or acoustic models and databases

• US 5,897,616 A
• Filed: 06/11/1997
• Issued: 04/27/1999
• Est. Priority Date: 06/11/1997
• Status: Expired due to Fees

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Kanevsky, Dimitri, Maes, Stephane Herman
• Primary Examiner(s): Hudspeth, David R.
• Assistant Examiner(s): MCFADDEN, SUSAN IRIS
• Application Number:
  US08/871,784
• Time in Patent Office: 685 Days
• Field of Search: 704/246, 704/247, 704/243, 704/272, 704/273, 704/275, 379/88, 379/67
• US Class Current: 704/246
• CPC Class Codes: G10L 17/22 Interactive procedures; Man-machine interfaces
• CPC Class Codes: G10L 17/24 the user being prompted to utter a password or a predefined phrase

Abstract

A method and apparatus for securing access to a service or facility employing automatic speech recognition, text-independent speaker identification, natural language understanding techniques and additional dynamic and static features. The method includes the steps of receiving and decoding speech containing indicia of the speaker such as a name, address or customer number; accessing a database containing information on candidate speakers; questioning the speaker based on the information; receiving, decoding and verifying an answer to the question; obtaining a voice sample of the speaker and verifying the voice sample against a model; generating a score based on the answer and the voice sample; and granting access if the score is equal to or greater than a threshold. Alternatively, the method includes the steps of receiving and decoding speech containing indicia of the speaker; generating a sub-list of speaker candidates having indicia substantially matching the speaker; activating databases containing information about the speaker candidates in the sub-list; performing voice classification analysis; eliminating speaker candidates based on the voice classification analysis; questioning the speaker regarding the information; eliminating speaker candidates based on the answer; and iteratively repeating prior steps until one speaker candidate (in which case the speaker is granted access), or no speaker candidate remains (in which case the speaker is not granted access).

Claims

1. A method of controlling access of a speaker to one of a service and a facility, the method comprising the steps of:

(a) receiving first spoken utterances of the speaker, the first spoken utterances containing indicia of the speaker;

(b) decoding the first spoken utterances;

(c) accessing a database corresponding to the decoded first spoken utterances, the database containing information attributable to a speaker candidate having indicia substantially similar to the speaker;

(d) querying the speaker with at least one question based on the information contained in the accessed database;

(e) receiving second spoken utterances of the speaker, the second spoken utterances being representative of at least one answer to the at least one question;

(f) decoding the second spoken utterances;

(g) verifying the accuracy of the decoded answer against the information contained in the accessed database serving as the basis for the question;

(h) taking a voice sample from the utterances of the speaker and processing the voice sample against an acoustic model attributable to the speaker candidate without requiring dependency on the decoded first and second spoken utterances;

(i) generating a score corresponding to the accuracy of the decoded answer and the closeness of the match between the voice sample and the model; and

(j) comparing the score to a predetermined threshold value and if the score is one of substantially equivalent to and above the threshold value, then permitting speaker access to one of the service and the facility.

2. The method of claim 1, further comprising the step of one of denying access to the speaker and repeating the process if the score is not substantially equivalent to and not above the threshold value.

3. The method of claim 1, wherein the acoustic model attributable to the speaker is not previously available and the method further comprising the steps of:

(a) querying the speaker with a plurality of questions based on the information contained in the accessed database; and

(b) building the acoustic model from voice samples taken from a plurality of answers provided by the speaker in response to the plurality of questions.

4. The method of claim 1, wherein the indicia in the first spoken utterances includes identifying indicia.

5. The method of claim 1, wherein the indicia in the first spoken utterances includes verifying indicia.

6. The method of claim 1, wherein at least a portion of the information contained in the database is one of acoustic and non-acoustic information.

7. The method of claim 1, wherein at least a portion of the information contained in the database is derived from spoken utterances provided by the speaker prior to the decoding step.

8. The method of claim 1, wherein at least a portion of the information contained in the database is derived from decoded spoken utterances provided by the speaker.

9. The method of claim 1, wherein at least a portion of the information in the database has static features.

10. The method of claim 1, wherein at least a portion of the information in the database has dynamic features.

11. The method of claim 1, wherein the sub-step of processing the voice sample against the acoustic model is performed by a text-independent speaker recognition technique.

12. The method of claim 1, further comprising the step of requerying the at least one question if the at least one answer is not accepted during the decoding step.

13. The method of claim 1, wherein the steps of the method of controlling access of the speaker forms a speech biometric.

14. The method of claim 1, wherein one of the database and the model may be built through pre-enrollment of the speaker.

15. The method of claim 1, wherein one of the database and the model may be one of built and updated automatically during the method of controlling access of the speaker to one of the service and the facility.

16. The method of claim 1, wherein at least a portion of the at least one answer provided by the speaker is in a form other than speech.

17. The method of claim 1, wherein the sub-step of processing the voice sample against the acoustic model is performed on a random question.

18. A method of controlling access of a speaker to one of a service and a facility from among a multiplicity of speaker candidates, the method comprising the steps of:

(a) receiving first spoken utterances of the speaker, the first spoken utterances containing indicia of the speaker;

(b) decoding the first spoken utterances;

(c) generating a sub-list of speaker candidates that substantially match the speakers decoded spoken utterances;

(d) activating databases respectively corresponding to the speaker candidates in the sub-list, the databases containing information respectively attributable to the speaker candidates;

(e) performing a voice classification analysis on voice characteristics of the speaker without requiring dependency on the decoded first spoken utterance;

(f) eliminating speaker candidates who do not substantially match these characteristics;

(g) querying the speaker with at least one question that is relevant to the information in the databases of speaker candidates remaining after the step (f);

(h) further eliminating speaker candidates based on the accuracy of the answer provided by the speaker in response to the at least one question;

(i) further performing the voice classification analysis on the voice characteristics from the answer provided by the speaker without requiring dependency on the decoded first spoken utterance;

(j) still further eliminating speaker candidates who do not substantially match these characteristics; and

(k) iteratively repeating steps (g) through (j) until one of one speaker candidate and no speaker candidates remain, if one speaker candidate remains then permitting the speaker access and if no speaker candidate remains then denying the speaker access.

19. The method of claim 18, wherein the first spoken utterances of the speaker are also representative of a password which is shared by a sub-set of the multiplicity of speaker candidates.

20. The method of claim 18, wherein the indicia in the first spoken utterances includes identifying indicia.

21. The method of claim 18, wherein the indicia in the first spoken utterances includes verifying indicia.

22. The method of claim 18, wherein at least a portion of the information contained in the database is one of acoustic and non-acoustic information.

23. The method of claim 18, wherein at least a portion of the information contained in the database is derived from spoken utterances provided by the speaker prior to the decoding step.

24. The method of claim 18, wherein at least a portion of the information contained in the database is derived from decoded spoken utterances provided by the speaker.

25. The method of claim 18, wherein at least a portion of the information in the database has static features.

26. The method of claim 18, wherein at least a portion of the information in the database has dynamic features.

27. The method of claim 18, wherein the steps of performing voice classification analysis are performed by a text-independent voice classification.

28. The method of claim 18, further comprising the step of requerying the at least one question if the at least one answer is not accepted during the decoding step.

29. The method of claim 18, wherein the steps of the method of controlling access of the speaker forms a speech biometric.

30. The method of claim 18, wherein one of the database and the model may be built through pre-enrollment of the speaker.

31. The method of claim 18, wherein one of the database and the model may be one of built and updated automatically during the method of controlling access of the speaker to one of the service and the facility.

32. The method of claim 18, wherein at least a portion of the at least on e answer provided by the speaker is in a form other than speech.

33. The method of claim 18, wherein the steps of performing voice classification analysis are performed on a random question.

34. Apparatus for controlling access of a speaker to one of a service and a facility, the apparatus comprising:

means for receiving first spoken utterances of the speaker, the first spoken utterances containing indicia of the speaker;

means for decoding the first spoken utterances;

means for accessing a database corresponding to the decoded first spoken utterances, the database containing information attributable to a speaker candidate having indicia substantially similar to the speaker;

means for querying the speaker with at least one question based on the information contained in the accessed database;

means for receiving second spoken utterances of the speaker, the second spoken utterances being representative of at least one answer to the at least one question;

means for decoding the second spoken utterances;

means for verifying the accuracy of the decoded answer against the information contained in the accessed database serving as the basis for the question;

means for taking a voice sample from the utterances of the speaker and processing the voice sample against an acoustic model attributable to the speaker candidate without requiring dependency on the decoded first and second spoken utterances;

means for generating a score corresponding to the accuracy of the decoded answer and the closeness of the match between the voice sample and the model; and

means for comparing the score to a predetermined threshold value and if the score is one of substantially equivalent to and above the threshold value, then permitting speaker access to one of the service and the facility.

35. Apparatus for controlling access of a speaker to one of a service and a facility from among a multiplicity of speaker candidates, the apparatus comprising:

means for receiving first spoken utterances of the speaker, the first spoken utterances containing indicia of the speaker;

means for decoding the first spoken utterances;

means for generating a sub-list of speaker candidates that substantially match the speakers decoded spoken utterances;

means for activating databases respectively corresponding to the speaker candidates in the sub-list, the databases containing information respectively attributable to the speaker candidates;

means for performing a voice classification analysis on voice characteristics of the speaker without requiring dependency on the decoded first spoken utterance;

means for eliminating speaker candidates who do not substantially match these characteristics;

means for querying the speaker with at least one question to the speaker that is relevant to the information in the databases of speaker candidates remaining after elimination by the eliminating means;

means for further eliminating speaker candidates based on the accuracy of the answer provided by the speaker in response to the at least one question;

means for further performing the voice classification analysis on the voice characteristics from the answer provided by the speaker without requiring dependency on the decoded first spoken utterance;

means for still further eliminating speaker candidates who do not substantially match these characteristics; and

means for iteratively repeating the querying and voice classification analysis procedures until one of one speaker candidate and no speaker candidate remains, if one speaker candidate remains then permitting the speaker access and if no speaker candidate remains then denying the speaker access.