Cognitive Intelligence Based Voice Authentication

• US 20180205726A1
• Filed: 01/17/2017
• Published: 07/19/2018
• Est. Priority Date: 01/17/2017
• Status: Active Application

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Chari, Suresh N., Singh, Kapil K.
• Granted Patent: US 10,277,590 B2
• Time in Patent Office: Days
• CPC Class Codes: G10L 15/26 Speech to text systems G10L15/08 takes precedence
• 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: H04L 63/0861 using biometrical features, e.g. fingerprint, retina-scan cryptographic mechanisms or cryptographic arrangements for entity authentication using biological data H04L9/3231
• CPC Class Codes: H04L 63/123 received data contents, e.g. message integrity

Abstract

Mechanisms are provided to detect a potentially fraudulent voice conversation. The mechanisms process a corpus of electronic information to extract a fraud feature representative of at least one fraudulent activity, receive a first voice input from a user, and convert the first voice input into a textual representation of the first voice input and a set of behavioral speech characteristics associated with the user. The mechanisms generate a speech model for the user based on the textual representation and the behavioral speech characteristics, receive a second voice input from an entity requesting access to resources associated with the user, and evaluate the second voice input based on the speech model for the user and the fraud feature. The mechanisms generate an output indicating whether or not the entity is the user based on results of the evaluation.

Claims

1. A method, in a data processing system comprising at least one processor and at least one memory, wherein the at least one memory comprises instructions which are executed by the at least one processor to specifically configure the at least one processor to detect a potentially fraudulent voice conversation, the method comprising:

processing, by the data processing system, a corpus of electronic information to extract a fraud feature representative of at least one fraudulent activity;

receiving, by the data processing system, a first voice input from a user;

converting, by the data processing system, the first voice input into a textual representation of the first voice input and a set of one or more behavioral speech characteristics associated with the user;

generating, by the data processing system, a speech model for the user based on the textual representation and the one or more behavioral speech characteristics;

receiving, by the data processing system, a second voice input from an entity requesting access to resources associated with the user;

evaluating, by the data processing system, the second voice input based on the speech model for the user and the fraud feature; and

generating, by the data processing system, an output indicating whether or not the entity is the user based on results of the evaluation.

2. The method of claim 1, wherein the one or more behavioral speech characteristics comprise a set of words or phrases that the user utilizes, equal to or more than a predetermined threshold number of times, when speaking with other persons.

3. The method of claim 2, wherein the set of words or phrases are identified in a predetermined portion of a conversation corresponding to one of a salutation at a beginning of the conversation or a final signature at a close of the conversation.

4. The method of claim 1, wherein the one or more behavioral speech characteristics comprise a pattern of pauses between words spoken by the user.

5. The method of claim 1, wherein the one or more behavioral speech characteristics comprise at least one of a pattern of inflections or an indicator of an accent utilized by the user.

6. The method of claim 1, wherein evaluating the second voice input comprises identifying a portion of the second voice input, having a feature that corresponds to the fraud feature, for processing based on the speech model.

7. The method of claim 1, wherein processing the corpus of electronic information to extract a fraud feature representative of at least one fraudulent activity comprises performing natural language processing on electronic natural language documents of the corpus to extract the fraud feature.

8. The method of claim 1, wherein the corpus of electronic information is a corpus associated with a domain corresponding to a protected or protective entity associated with the data processing system, and wherein at least one electronic document in the corpus comprises information describing aspects of fraudulent activity from which the fraud feature is extracted, and wherein the fraud feature specifies at least one manner by which a perpetrator of the fraudulent activity communicates with another entity.

9. The method of claim 1, further comprising training a cognitive system based on the speech model to recognize speech patterns of the user based on the one or more behavioral speech characteristics, wherein evaluating the second voice input comprises:

converting the second voice input to a second textual representation and second set of one or more behavioral speech characteristics;

identifying a portion of the second textual representation and corresponding second set of one or more behavioral speech characteristics based on the fraud feature; and

comparing the portion of the second textual representation and corresponding second set of one or more behavioral speech characteristics to the one or more behavioral speech characteristics in the speech model to generate a measure of confidence that the portion of the second textual representation is part of a fraudulent activity.

10. The method of claim 9, wherein comparing the portion of the second textual representation and corresponding second set of one or more behavioral speech characteristics to the one or more behavioral speech characteristics in the speech model to generate a measure of confidence comprises:

scoring features of the portion of the second textual representation based on the comparison to generate a fraud score; and

comparing the fraud score to a fraud threshold value, wherein generating the output comprises generating a first output indicating fraudulent activity in response to the fraud score equaling or exceeding the fraud threshold value or generating a second output indicating no fraudulent activity in response to the fraud score not equaling or exceeding the fraud threshold value.

11. A computer program product comprising a computer readable storage medium having a computer readable program stored therein, wherein the computer readable program, when executed on a computing device, causes the computing device to be specifically configured to detect a potentially fraudulent voice conversation at least by:

processing a corpus of electronic information to extract a fraud feature representative of at least one fraudulent activity;

receiving a first voice input from a user;

converting the first voice input into a textual representation of the first voice input and a set of one or more behavioral speech characteristics associated with the user;

generating a speech model for the user based on the textual representation and the one or more behavioral speech characteristics;

receiving a second voice input from an entity requesting access to resources associated with the user;

evaluating the second voice input based on the speech model for the user and the fraud feature; and

generating an output indicating whether or not the entity is the user based on results of the evaluation.

12. The computer program product of claim 11, wherein the one or more behavioral speech characteristics comprise a set of words or phrases that the user utilizes, equal to or more than a predetermined threshold number of times, when speaking with other persons.

13. The computer program product of claim 12, wherein the set of words or phrases are identified in a predetermined portion of a conversation corresponding to one of a salutation at a beginning of the conversation or a final signature at a close of the conversation.

14. The computer program product of claim 11, wherein the one or more behavioral speech characteristics comprise at least one of a pattern of pauses between words spoken by the user, a pattern of inflections used by the user, or an indicator of an accent utilized by the user.

15. The computer program product of claim 11, wherein the computer readable program further causes the computing device to evaluate the second voice input at least by identifying a portion of the second voice input, having a feature that corresponds to the fraud feature, for processing based on the speech model.

16. The computer program product of claim 11, wherein the computer readable program further causes the computing device to process the corpus of electronic information to extract a fraud feature representative of at least one fraudulent activity at least by performing natural language processing on electronic natural language documents of the corpus to extract the fraud feature.

17. The computer program product of claim 11, wherein the corpus of electronic information is a corpus associated with a domain corresponding to a protected or protective entity associated with the data processing system, and wherein at least one electronic document in the corpus comprises information describing aspects of fraudulent activity from which the fraud feature is extracted, and wherein the fraud feature specifies at least one manner by which a perpetrator of the fraudulent activity communicates with another entity.

18. The computer program product of claim 11, wherein the computer readable program further causes the computing device to train a cognitive system based on the speech model to recognize speech patterns of the user based on the one or more behavioral speech characteristics, and wherein the computer readable program further causes the computing device to evaluate the second voice input at least by:

converting the second voice input to a second textual representation and second set of one or more behavioral speech characteristics;

identifying a portion of the second textual representation and corresponding second set of one or more behavioral speech characteristics based on the fraud feature; and

comparing the portion of the second textual representation and corresponding second set of one or more behavioral speech characteristics to the one or more behavioral speech characteristics in the speech model to generate a measure of confidence that the portion of the second textual representation is part of a fraudulent activity.

19. The computer program product of claim 18, wherein the computer readable program further causes the computing device to compare the portion of the second textual representation and corresponding second set of one or more behavioral speech characteristics to the one or more behavioral speech characteristics in the speech model to generate a measure of confidence at least by:

scoring features of the portion of the second textual representation based on the comparison to generate a fraud score; and

comparing the fraud score to a fraud threshold value, wherein generating the output comprises generating a first output indicating fraudulent activity in response to the fraud score equaling or exceeding the fraud threshold value or generating a second output indicating no fraudulent activity in response to the fraud score not equaling or exceeding the fraud threshold value.

20. An apparatus comprising:

a processor; and

a memory coupled to the processor, wherein the memory comprises instructions which, when executed by the processor, cause the processor to be specifically configured to detect a potentially fraudulent voice conversation at least by;

processing a corpus of electronic information to extract a fraud feature representative of at least one fraudulent activity;

receiving a first voice input from a user;

converting the first voice input into a textual representation of the first voice input and a set of one or more behavioral speech characteristics associated with the user;

generating a speech model for the user based on the textual representation and the one or more behavioral speech characteristics;

receiving a second voice input from an entity requesting access to resources associated with the user;

evaluating the second voice input based on the speech model for the user and the fraud feature; and

generating an output indicating whether or not the entity is the user based on results of the evaluation.