Speaker verification method

• US 7,630,895 B2
• Filed: 01/12/2005
• Issued: 12/08/2009
• Est. Priority Date: 01/21/2000
• Status: Expired due to Fees

• Current Assignee: Nuance Communications, Inc. (Microsoft Corporation)
• Original Assignee: AT&T Intellectual Property I LP (AT&T, Inc.)
• Inventors: Ratasuk, Rapeepat, Bossemeyer, Jr., Robert Wesley
• Primary Examiner(s): Opsasnick, Michael N
• Application Number:
  US11/033,697
• Publication Number:
  US 20050143996A1
• Time in Patent Office: 1,791 Days
• Field of Search: 704/231, 704/246, 704/255
• US Class Current: 704/246
• CPC Class Codes: G10L 17/06 Decision making techniques; Pattern matching strategies
• CPC Class Codes: G10L 17/08 Use of distortion metrics or a particular distance between probe pattern and reference templates

Abstract

A speaker verification method consist of the following steps: (1) generating a code book (42) covering a number of speakers having a number of training utterances for each of the speakers; (2) receiving a number of test utterances (44) from a speaker; (3) comparing (46) each of the test utterances to each of the training utterances for the speaker to form a number of decisions, one decision for each of the number of test utterances; (4) weighting each of the decisions (48) to form a number of weighted decisions; and (5) combining (50) the plurality of weighted decisions to form a verification decision (52).

Claims

1. A speaker verification method comprising:

performing operations executed by a speaker verification machine, the operations comprising;

comparing a plurality of different test utterances to a plurality of training utterances for a speaker to form a plurality of preliminary verification decisions, one preliminary verification decision for each of the plurality of different test utterances;

weighting each of the plurality of preliminary verification decisions based on a historical error rate corresponding to a respective one of the test utterances; and

combining the weighted preliminary verification decisions to form a verification decision.

2. The method of claim 1, wherein the one preliminary verification decision is either a true or a false decision.

3. The method of claim 1, further comprising generating a code book covering a plurality of speakers having the plurality of training utterances for each of the plurality of speakers, wherein a first utterance of the plurality of training utterances is not the same as a second utterance of the plurality of training utterances.

4. The method of claim 1, wherein weighting each of the plurality of preliminary verification decisions includes determining a historical probability of false alarm for each of the plurality of different test utterances.

5. The method of claim 1, wherein weighting each of the plurality of preliminary verification decisions includes evaluating a quality of the preliminary verification decision for each of the plurality of preliminary verification decisions.

6. The method of claim 1, further comprising:

receiving a plurality of input utterances;

segmenting the plurality of input utterances into voiced sounds and unvoiced sounds; and

extracting features from the voiced sounds to form the plurality of different test utterances.

7. The method of claim 1, further comprising generating a code book associated with a plurality of speakers and including the plurality of training utterances.

8. The method of claim 7, wherein generating the code book comprises:

receiving a sample test utterance;

segmenting the sample test utterance into voiced sounds and unvoiced sounds;

extracting features from the voiced sounds; and

storing the features as one of the training utterances.

9. The method of claim 1, further comprising selecting a male variance vector to weight each of the plurality of preliminary verification decisions based on whether the speaker is a male.

10. The method of claim 9, further comprising determining whether the speaker purports to be a male.

11. A computer readable medium having instructions stored thereon that, when executed, cause a machine to:

compare a plurality of different test utterances to a plurality of training utterances for the speaker to form a plurality of preliminary verification decisions, one preliminary verification decision for each of the plurality of different test utterances;

weight each of the plurality of preliminary verification decisions based on a historical error rate corresponding to a respective one of the test utterances; and

combine the weighted preliminary verification decisions to form a verification decision.

12. The computer readable medium of claim 11 having instructions stored thereon that, when executed, cause the machine to:

receive a plurality of input utterances;

segment the plurality of input utterances into voiced sounds and unvoiced sounds; and

extract features from the voiced sounds to form the plurality of different test utterances.

13. The computer readable medium of claim 11 having instructions stored thereon that, when executed, cause the machine to generate a code book covering a plurality of speakers having the plurality of training utterances for each of the plurality of speakers, wherein a first utterance of the plurality of training utterances is not the same as a second utterance of the plurality of training utterances.

14. The computer readable medium of claim 11 having instructions stored thereon that, when executed, cause the machine to generate a code book associated with a plurality of speakers and including the plurality of training utterances.

15. The computer readable medium of claim 14 having instructions stored thereon that, when executed, cause the machine to generate the code book by:

receiving a sample test utterance;

segmenting the sample test utterance into voiced sounds and unvoiced sounds;

extracting features from the voiced sounds; and

storing the features as one of the training utterances.

16. The computer readable medium of claim 11 having instructions stored thereon that, when executed, cause the machine to select a male variance vector to weight each of the plurality of preliminary verification decisions based on whether the speaker is a male.

17. The computer readable medium of claim 16 having instructions stored thereon that, when executed, cause the machine to determine whether the speaker purports to be a male.

18. A computer readable medium having instructions stored thereon that, when executed, cause a machine to:

determine whether a speaker is male;

when the speaker is male, use a male variance vector to determine a weighted Euclidean distance between each of a plurality of test utterances and a plurality of training utterances;

form a preliminary verification decision for each of the plurality of test utterances to form a plurality of preliminary verification decisions;

combine the plurality of preliminary verification decisions to form a verification decision; and

weight each of the plurality of preliminary verification decisions based on a historical error rate corresponding to a respective one of the test utterances.

19. The computer readable medium of claim 18 having instructions stored thereon that, when executed, cause the machine to:

separate a plurality of speakers into a male group and a female group; and

determine the male variance vector based on the male group.

20. The computer readable medium of claim 18 having instructions stored thereon that, when executed, cause the machine to retrieve the male variance vector from a code book including the plurality of training utterances.

21. The computer readable medium of claim 18 having instructions stored thereon that, when executed, cause the machine to:

receive a plurality of input utterances;

segment the plurality of input utterances into voiced sounds and unvoiced sounds; and

extract features from the voiced sounds to form the plurality of test utterances.