Multi-resolution system and method for speaker verification

• US 6,272,463 B1
• Filed: 03/03/1999
• Issued: 08/07/2001
• Est. Priority Date: 03/03/1998
• Status: Expired due to Term

• Current Assignee: Nuance Communications, Inc. (Microsoft Corporation)
• Original Assignee: Lernout & Hauspie Speech Products NV (Intel Corporation)
• Inventors: Lapere, Martine
• Primary Examiner(s): Dorvil, Richemond
• Assistant Examiner(s): AZAD, ABUL K
• Application Number:
  US09/261,271
• Time in Patent Office: 888 Days
• Field of Search: 704/246, 704/247, 704/248, 704/249, 704/250, 704/273, 704/251, 704/253, 704/254, 704/255, 704/243, 704/260, 704/257, 704/256
• US Class Current: 704/248
• CPC Class Codes: G10L 15/065 Adaptation
• CPC Class Codes: G10L 17/04 Training, enrolment or model building

Abstract

A method is given for generating a speaker-dependant model of an utterance that has at least one occurrence. The method includes generating an initial model, having a first resolution, that encodes each of the occurrences of the utterance; and generating at least one additional speaker-specific model, having a different resolution from that of the initial model, of all occurrences of the utterance.

Claims

1. A multi-resolution model speaker verification system having at least one authorized user, the system comprising:

a database of high resolution word models containing word models associated with each authorized user;

a verification phrase generator in communication with the database of high resolution word models that generates a verification phrase when a user seeks to be verified as a specific authorized user, the verification phrase containing at least one key word having a model in the database of high resolution models, and at least one other word;

a verification phrase model generator in communication with the verification phrase generator that generates a comparison model of the verification phrase;

a user access device in communication with the verification phrase generator that communicates the verification phrase to the user and that allows a spoken response phrase having words to be provided by the user;

an anti-synthesis comparator that checks the response phrase for acoustic similarity and continuity to determine if the response phrase was produced by natural speech; and

a verification engine in communication with the user access device, the anti-synthesis comparator, and the verification phrase model generator that verifies the user as the specific authorized user when;

(i) the response phrase matches the comparison model,

(ii) for the at least one key word, the models in the database of high resolution word models that are associated with the specific authorized user are within a threshold distance of corresponding words in the response phrase, and

(iii) the anti-synthesis comparator determines that the response phrase was produced by natural speech.

2. A system as in claim 1, wherein the at least one other word in the verification phrase has a high resolution word model from the database of high resolution word models.

3. A system as in claim 1, wherein the at least one other word in the verification phrase has a low resolution word model from a database of low resolution word models.

4. A system as in claim 1, wherein the user access device includes a text-to-speech module that communicates the verification phrase to the user.

5. A system as in claim 1, wherein the user access device includes an automatic speech recognition module that allows a response phrase to be provided by the user.

6. A computer network including the speaker verification system of claim 1 to limit network access to authorized users.

7. A method of speaker verification comprising:

establishing a database of high resolution word models containing word models associated with each authorized user;

generating a verification phrase with a verification phrase generator in communication with the database of high resolution word models when a user seeks to be verified as a specific authorized user, the verification phrase containing at least one key word having a model in the database of high resolution models, and at least one other word;

generating a comparison model of the verification phrase with a verification phrase model generator in communication with the verification phrase generator;

communicating the verification phrase to the user with a user access device in communication with the grammar compiler that allows a spoken response phrase having words to be provided by the user;

checking the words in the response phrase for acoustic similarity and continuity with an anti-synthesis comparator to determine if the response phrase was produced by natural speech; and

verifying the user as the specific authorized user with a verification engine in communication with the user access device and the verification model generator when;

(i) the response phrase matches the comparison model,

(ii) for the at least one key word, the models in the database of high resolution word models that are associated with the specific authorized user are within a threshold distance of corresponding words in the response phrase, and

(iii) the response phrase has been determined by the anti-synthesis comparator to be produced by natural speech.

8. A method as in claim 7, wherein the at least one other word in the verification phrase has a high resolution word model from the database of high resolution word models.

9. A method as in claim 7, wherein the at least one other word in the verification phrase has a low resolution word model from a database of low resolution word models.

10. A method as in claim 7, wherein the user access device includes a text-to-speech module that communicates the verification phrase to the user.

11. A method as in claim 7, wherein the user access device includes an automatic speech recognition module that allows a response phrase to be provided by the user.

12. A computer network employing the speaker verification method of claim 7 to limit network access to authorized users.