Adaptive voice authentication system and method

• US 9,940,934 B2
• Filed: 03/09/2016
• Issued: 04/10/2018
• Est. Priority Date: 11/18/2015
• Status: Active Grant

• Current Assignee: Uniphore Technologies, Inc. (Uniphore Software Systems Pvt Ltd.)
• Original Assignee: Uniphore Software Systems Co.
• Inventors: Sachdev, Umesh
• Primary Examiner(s): COLUCCI, MICHAEL C
• Application Number:
  US15/064,740
• Publication Number:
  US 20170140760A1
• Time in Patent Office: 762 Days
• Field of Search: 704246, 704231, 704233, 704236, 705 35, 705 44, 713166, 713168, 726 17, 726 22, 726 26
• CPC Class Codes: G10L 17/02 Preprocessing operations, e.g. segment selection; Pattern representation or modelling, e.g. based on linear discriminant analysis [LDA] or principal components; Feature selection or extraction
• CPC Class Codes: G10L 17/04 Training, enrolment or model building
• CPC Class Codes: G10L 17/06 Decision making techniques; Pattern matching strategies
• CPC Class Codes: G10L 25/15 the extracted parameters being formant information
• CPC Class Codes: G10L 25/24 the extracted parameters being the cepstrum
• CPC Class Codes: G10L 25/75 for modelling vocal tract parameters

Abstract

An adaptive voice authentication system is provided. The adaptive voice authentication system includes an adaptive module configured to compare a feature quality index of the plurality of authentication features and the plurality of enrollment features and dynamically replace and store one or more enrollment features with one or more authentication features to form a plurality of updated enrollment features. The adaptive module is configured to generate an updated enrollment voice print model from the plurality of the updated enrollment features. The adaptive module is further configured to compare the updated enrollment voice print model with the previously stored enrollment voice print model and dynamically update the previously stored enrollment voice print model with the updated enrollment voice print model based on a model quality index.

Claims

1. An adaptive voice authentication system comprising:

a memory configured to store computer-readable instructions; and

a processor configured to execute the computer-readable instructions to,

receive an enrolment voice sample of a user and an authentication voice sample of the user;

extract a plurality of enrolment features from the enrolment voice sample and a plurality of authentication features from the authentication voice sample;

generate an enrolment voice print model from the plurality of enrolment features and an authentication voice print model from the plurality of authentication features;

receive the authentication voice print model and authenticate the user based on the enrolment voice print model;

store the plurality of enrolment features, the plurality of authentication features, the enrolment voice print model, and the authentication voice print model;

compare a feature quality index of the plurality of authentication features and the plurality of enrolment features each time an authentication of the user is initiated, wherein the feature quality index is determined based on one or more of a signal to noise ratio, dynamic range level, loudness, and consistency of the plurality of enrolment features and the plurality of authentication features;

dynamically update one or more enrolment features with one or more corresponding authentication features to form one or more updated enrolment features in response to the feature quality index being greater than or equal to a feature quality threshold value, the one or more corresponding authentication features being of a higher quality than the one or more enrolment features;

generate an updated enrolment voice print model using the one or more updated enrolment features;

compare the updated enrolment voice print model with the stored enrolment voice print model using another voice sample of the user as input; and

dynamically update the stored enrolment voice print model with the updated enrolment voice print model, in response to the updated enrolment voice print model having a higher likelihood score for the user than the stored enrolment voice print model, based on a model quality index, wherein the model quality index is determined based on one or more of a signal to noise ratio, dynamic range level, loudness, and consistency of the enrolment voice print model and the authentication voice print model.

2. The adaptive voice authentication system of claim 1, wherein the plurality of enrolment features and the plurality of authentication features comprise vocal tract shape and length, short term and long term energy, fundamental pitch, formant features, speaking rate, prosody features, language and accent, nasality, cepstrum, emotional state or combinations thereof.

3. The adaptive voice authentication system of claim 1, the processor is further configured to execute the computer-readable instructions to improve the enrolment voice sample and the authentication voice sample by filtering a plurality of distortion elements.

4. The adaptive voice authentication system of claim 3, wherein filtering the plurality of distortion elements includes employing one or more filtering operations comprising clipping, smoothening, and amplifying the enrolment voice sample and the authentication voice sample to generate an improved enrolment voice sample and an improved authentication voice sample.

5. The adaptive voice authentication system of claim 1, wherein the processor is configured to execute the computer-readable instructions to compare each authentication feature with all of the plurality of enrolment features each time an authentication of the user is initiated.

6. The adaptive voice authentication system of claim 1, wherein the processor is configured to execute the computer-readable instructions to compare the model quality index each time the one or more enrolment features are updated with the one or more corresponding authentication features of a higher quality to form the one or more updated enrolment features.

7. The adaptive voice authentication system of claim 1, wherein the processor is configured to execute the computer-readable instructions to store the feature quality index and the model quality index.

8. A method for dynamically updating one or more enrolment features and an enrolment voice print model, the method comprising:

receiving an enrolment voice sample of a user and an authentication voice sample of the user;

extracting a plurality of enrolment features from the enrolment voice sample and a plurality of authentication features from the authentication voice sample;

generating an enrolment voice print model from the plurality of enrolment features and an authentication voice print model from the plurality of authentication features;

authenticating the user based on the enrolment voice print model;

storing the plurality of enrolment features, the plurality of authentication features, the enrolment voice print model, and the authentication voice print model;

comparing a feature quality index of the plurality of authentication features and the plurality of enrolment features each time an authentication of the user is initiated, wherein the feature quality index is determined based on one or more of a signal to noise ratio, dynamic range level, loudness, and consistency of the plurality of enrolment features and the plurality of authentication features;

dynamically updating one or more enrolment features with one or more corresponding authentication features to form one or more updated enrolment features in response to the feature quality index being greater than or equal to a feature quality threshold value, the one or more corresponding authentication features being of a higher quality than the one or more enrolment features;

generating an updated enrolment voice print model using the one or more updated enrolment features;

comparing the updated enrolment voice print model with the stored enrolment voice print model using another voice sample of the user as input; and

dynamically updating the stored enrolment voice print model with the updated enrolment voice print model, in response to the updated enrolment voice print model having a higher likelihood score for the user than the stored enrolment voice print model, based on a model quality index, wherein the model quality index is determined based on one or more of a signal to noise ratio, dynamic range level, loudness, and consistency of the enrolment voice print model and the authentication voice print model.

9. The method of claim 8, wherein the plurality of enrolment features and the plurality of authentication features comprise vocal tract shape and length, short term and long term energy, fundamental pitch, formant features, speaking rate, prosody features, language and accent, nasality, cepstrum, emotional state or combinations thereof.

10. The method of claim 8, further comprising:

improving the enrolment voice sample and the authentication voice sample by filtering a plurality of distortion elements, wherein filtering the plurality of distortion elements includes employing one or more filtering operations comprising clipping, smoothening, and amplifying the enrolment voice sample and the authentication voice sample to generate an improved enrolment voice sample and an improved authentication voice sample.

11. The method of claim 8, further comprising comparing each authentication feature with all of the plurality of enrolment features each time an authentication of the user is initiated.

12. The method of claim 8, wherein comparing the model quality index is performed each time the one or more enrolment features are updated with the one or more corresponding authentication features of a higher quality to form the one or more updated enrolment features.

13. The method of claim 8, further comprising storing the feature quality index and the model quality index.