System and method for speaker verification using short utterance enrollments

• US 20060178885A1
• Filed: 02/07/2005
• Published: 08/10/2006
• Est. Priority Date: 02/07/2005
• Status: Active Grant

• Current Assignee: Hitachi, Ltd.
• Original Assignee: Hitachi, Ltd.
• Inventors: Tavares, Clifford
• Granted Patent: US 7,490,043 B2
• Time in Patent Office: Days
• US Class Current: 704/246
• CPC Class Codes: G10L 17/08 Use of distortion metrics or a particular distance between probe pattern and reference templates

Abstract

A system and method are disclosed for verifying an identity of a speaker. In accordance with one embodiment, feature vectors are received that were extracted from an utterance made by a speaker claiming an identity. Dissimilarity may be measured between the feature vectors and a codebook associated with a version of the utterance known to be made by the identity. The utterance may be further analyzed to ascertain information about repeating occurrences of the feature vectors in the utterance. The information about repeating occurrences of feature vectors occurring in the utterance may be compared to information about repeating occurrences of feature vectors in a version of the utterance known to be made by the claimed identity. Based on the comparison, a penalty may be assigned to the measured dissimilarity. Using the measured dissimilarity modified by the assigned penalty, a determination may be made as to whether to accept the speaker as the identity.

Claims

1. A method for verifying an identity of a speaker, comprising:

receiving feature vectors extracted from an utterance made by a speaker claiming an identity;

measuring dissimilarity between the feature vectors and a codebook associated with a version of the utterance known to be made by the identity;

analyzing the utterance to ascertain information about repeating occurrences of the feature vectors in the utterance;

comparing the information about repeating occurrences of feature vectors occurring in the utterance with information about repeating occurrences of feature vectors in a version of the utterance known to be made by the claimed identity;

assigning a penalty to the measured dissimilarity based on the comparison; and

determining whether to accept the speaker as the identity using the measured dissimilarity modified by the assigned penalty.

2. The method of claim 1, wherein vector quantization is utilized to measure dissimilarity between the feature vectors and the codebook.

3. The method of claim 1, wherein the utterance has a duration between about 0.1 seconds and about 5 seconds.

4. The method of claim 1, wherein the utterance comprises a multi-syllabic utterance.

5. The method of claim 1, wherein if the number of repeating occurrences for any of the feature vectors of the utterance exceeds a maximum number, rejecting the speaker as the identity.

6. The method of claim 5, wherein an additional penalty is assigned to the dissimilarity if any of the feature vectors of the utterance is determined to have a number of repeating occurrences exceeding the maximum number.

7. The method of claim 5, wherein the maximum number for a given feature vector is obtained by analyzing a plurality of utterances made by a plurality of speakers to identify the utterance of the plurality of utterances having the largest number of repeating occurrences of the given feature vector.

8. The method of claim 1, wherein the assigned penalty comprises a penalty for each of the feature vectors of the utterance, wherein the assigned penalty for each feature vector is based on a difference between a number of repeating occurrences of the respective feature vector of the utterance and a number of repeating occurrences of the corresponding feature vector of the version of the utterance know to be made by the identity.

9. The method of claim 8, wherein the assigned penalty for given feature vector is adjusted based on the degree of difference between the number of repeating occurrences of the respective feature vector of the utterance and the number of repeating occurrences of the corresponding feature vector of the version of the utterance know to be made by the identity.

10. The method of claim 8, wherein the assigned penalty for each feature vector is adjusted to account for operational characteristics of a device used to capture the utterance by the speaker.

11. The method of claim 8, wherein no penalty is assigned to a given feature vector if the difference between the number of repeating occurrences of the respective feature vector of the utterance and the number of repeating occurrences of the corresponding feature vector of the version of the utterance know to be made by the identity is determined to be less than an expected difference of repeating occurrences occurring due to expected changes in voice.

12. The method of claim 8, wherein the assigned penalty for a given feature vector is reduced if the difference between the number of repeating occurrences of the respective feature vector of the utterance and the number of repeating occurrences of the corresponding feature vector of the version of the utterance know to be made by the identity is determined to be less than a predefined value.

13. The method of claim 1, wherein the determining whether to accept the speaker as the identity further comprises comparing the measured dissimilarity modified by the assigned penalty to a valid-imposter model associated with the utterance.

14. The method of claim 1, wherein the utterance comprises a plurality of frames, and wherein analyzing the utterance further comprises identifying the feature vectors occurring in each frame, counting the instances that each different feature vector of the utterance occurs in all of the frames to obtain a sum of repeating occurrences of each feature vector, and dividing each sum by a total number of repeating occurrences occurring in the utterance.

15. A system for verifying an identity of a speaker, comprising:

logic that receives feature vectors extracted from an utterance made by a speaker claiming an identity;

logic that measures dissimilarity between the feature vectors and a codebook associated with a version of the utterance known to be made by the identity;

logic that analyzes the utterance to ascertain information about repeating occurrences of the feature vectors in the utterance;

logic that compares the information about repeating occurrences of feature vectors occurring in the utterance with information about repeating occurrences of feature vectors in a version of the utterance known to be made by the claimed identity;

logic that assigns a penalty to the measured dissimilarity based on the comparison; and

logic that determines whether to accept the speaker as the identity using the measured dissimilarity modified by the assigned penalty.

16. The system of claim 15, wherein vector quantization is utilized to measure dissimilarity between the feature vectors and the codebook.

17. The system of claim 15, wherein the assigned penalty is adjusted to account for operational characteristics of a device used to capture the utterance by the speaker.

18. The system of claim 15, wherein the logic that determines whether to accept the speaker as the identity further comprises logic that compares the measured dissimilarity modified by the assigned penalty to a valid-imposter model associated with the utterance.

19. A method, comprising:

obtaining an utterance comprising a plurality of frames and having a plurality of feature vectors;

identifying the feature vectors present in each frame;

tabulating the presence of feature vectors by frame for the whole utterance;

identifying the number of instances each feature vector is repeated in the utterance;

calculating a total sum of all repeating instances in the utterance;

dividing the number of repeats for each feature vector by the total sum to obtain an averaged value for each feature vector; and

storing the information about the number of repeats for each feature vector in a reference log associated with the utterance.

20. The method of claim 19, further comprising examining the reference logs of a plurality of utterances made by a plurality of speakers to identify a set of feature vectors comprising all of the different feature vectors present in the reference logs, and for each different feature vector, identifying the largest number of repeat instances for that feature vector in a single reference log, and generating a global reference log that indicates the largest number of repeat instances for every feature vector.