SYSTEMS, METHODS, AND MEDIA FOR DISAMBIGUATING CALL DATA TO DETERMINE FRAUD

• US 20120263285A1
• Filed: 05/29/2012
• Published: 10/18/2012
• Est. Priority Date: 04/21/2005
• Status: Active Grant

• Current Assignee: Verint Americas Incorporated (Verint Systems Incorporated)
• Original Assignee: Verint Americas Incorporated (Verint Systems Incorporated)
• Inventors: Rajakumar, Anthony, Guerra, Lisa, Zeppenfeld, Torsten, Vyas, Vipul
• Granted Patent: US 9,113,001 B2
• Time in Patent Office: Days
• US Class Current: 379/189
• CPC Class Codes: G06F 21/32 using biometric data, e.g. fingerprints, iris scans or voiceprints
• CPC Class Codes: G06Q 20/4016 involving fraud or risk level assessment in transaction processing
• CPC Class Codes: G10L 17/00 Speaker identification or verification techniques
• CPC Class Codes: G10L 17/22 Interactive procedures; Man-machine interfaces
• CPC Class Codes: H04M 15/47 Fraud detection or prevention means
• CPC Class Codes: H04M 2201/40 using speech recognition speech recognition per se G10L15/00
• CPC Class Codes: H04M 2201/41 using speaker recognition speaker recognition per se G10L17/00
• CPC Class Codes: H04M 2203/6027 Fraud preventions
• CPC Class Codes: H04M 2215/0148 Fraud detection or prevention means
• CPC Class Codes: H04M 3/2281 Call monitoring, e.g. for law enforcement purposes; Call tracing; Detection or prevention of malicious calls
• CPC Class Codes: H04M 3/385 using speech signals
• CPC Class Codes: H04M 3/42059 Making use of the calling party identifier

Abstract

Systems, methods, and media for disambiguating call data are provided herein. Some exemplary methods include receiving, via a fraud notification system, notification of a fraud event associated with a customer account, the fraud event comprising a time stamp, determining, via a call selection module, unique voice samples or models from call events obtained within a time frame that is temporally proximate the fraud event, and generating a timeline presentation that includes each unique voice sample or model identified in the call events based upon a time stamp associated with the call events.

Claims

1. A method for disambiguating call data, the method comprising:

receiving, via a fraud notification module, notification of a fraud event including a customer account identifier and a fraud time stamp;

determining a time frame that is proximate the fraud time stamp;

collecting, using a call selection module, call events associated with the customer account identifier that occur during the determined time frame, each call event including a unique call event identifier, a voice sample, and a call event time stamp;

identifying, using a voice module, a first call event belonging to a first speaker, and a second call event belonging to a second speaker; and

generating, using a user interface module, a timeline presentation that includes the first call event and call event timestamp and an identification of the first voice sample as belonging to the first speaker, the second call event and call event timestamp and an identification of the second voice sample as belonging to the second speaker.

2. The method according to claim 1, wherein the order of the first and second call events on the time line is based on the call first and second call event time stamps.

3. The method according to claim 1, further comprising identifying, using the voice module, a third call event and timestamp associated with a third voice sample belonging to the first speaker.

4. The method according to claim 3, wherein the timeline further includes the third call event and timestamp and an identification of the third voice sample as belonging to the first speaker.

5. The method according to claim 1, further comprising identifying, using the voice module, a third call event and timestamp associated with a third voice sample belonging to a third speaker.

6. The method according to claim 5, wherein the timeline further includes the third call event and timestamp and an identification of the third voice sample as belonging to the third speaker.

7. The method according to claim 1, further comprising:

comparing the first and second voice sample to a voice model for a customer associated with the customer account identifier;

annotating the first call event based on a match between the first voice sample and the voice model of the customer; and

annotating the second call event based on a match between the second voice sample and the voice model of the customer.

8. The method according to claim 7, further comprising:

generating a first confidence estimate based on the comparison of the first voice sample to the voice model for the customer, the first confidence estimate representing a likelihood that the first voice sample corresponds to the voice model of the customer and using the first confidence estimate for annotating the first call event; and

generating a second confidence estimate based on the comparison of the second voice sample to the voice model for the customer, the second confidence estimate representing a likelihood that the second voice sample corresponds to the voice model of the customer and using the second confidence estimate for annotating the second call event.

9. The method according to claim 8, further comprising:

storing a first voice model extracted from the first voice sample in a whitelist based on the first confidence estimate; and

storing a second voice model extracted from the second voice sample in a whitelist based on the second confidence estimate.

10. The method according to claim 7, further comprising:

generating a first confidence estimate representing a likelihood that the first voice sample corresponds to the voice model of the customer, and based on the first confidence estimate comparing the first voice sample to voice models in a fraudster database that includes voice models of known fraudsters; and

generating a second confidence estimate representing a likelihood that the second voice sample corresponds to the voice model of the customer and based on the second confidence estimate comparing the second voice sample to voice models in the fraudster database.

11. The method according to claim 10, further comprising:

storing a first voice model extracted from the first voice sample based on the comparison of the first voice sample to voice models in the fraudster database; and

storing a second voice model extracted from the second voice sample based on the comparison of the second voice sample to voice models in the fraudster database.

12. The method according to claim 11, further comprising confirming that the stored first voice model is associated with the fraudster by comparing non-audio data associated with the first call event, wherein the non-audio data includes any of an automatic number identification, caller identification information, an international mobile equipment identity number, a given name, the timestamp associated with the call event, keywords included in the call event, and combinations thereof.

13. The method according to claim 1, further comprising diarizing the first and second call events to remove agent voice segments by:

detecting voice segments for an agent in the first and second voice samples; and

removing voice segments of belonging to the agent from the first and second voice samples.

14. The method according to claim 13, wherein removed voice segments are omitted from the timeline.

15. The method according to claim 1, wherein each call event comprises a screening identification for associating the call event with the customer account.

16. The method according to claim 1, further comprising grouping call events for substantially similar voice samples into a list, the list comprising a timestamp for each voice sample.

17. The method according to claim 1, further comprising providing, via a graphical user interface, a timeline that includes a first icon representing the first voice sample and a second icon representing the second voice sample.

18. A system for disambiguating call data, the system comprising:

a memory for storing executable instructions that disambiguate call data;

a processor that executes the instructions;

a fraud notification module stored in memory and executable by the processor to receive notification of a fraud event including a customer account identifier and a fraud time stamp;

a call selection module stored in memory and executable by the processor to determine a time frame that is proximate the fraud time stamp, and collect call events associated with the customer account that occur during the determined time frame, each call event including a unique identifier, a voice sample, and a call event time stamp;

a voice module stored in memory and executable by the processor to identify a first voice sample belonging to a first voice from collected call events and a second voice sample belonging to a second voice from collected call events; and

a scoring module stored in memory and executable by the processor to a determine for the first and second voice a first and second voice score respectively, the first and second voice score each representing a probability of the respective voice being a fraudster.

19. The system according to claim 18, further comprising a user interface module stored in memory and executable by the processor to generate a visual timeline that includes the identification of the first voice sample as belonging to the first voice, and the identification of the second voice sample as belonging to the second speaker.

20. The system according to claim 18, wherein the voice module is further executable by the processor to extract a first voice model based on the first voice sample and a second voice model based on the second voice sample.

21. The system according to claim 20, wherein the diarization module is further executable to group a first set of call events including voice models substantially matching the first voice model and to group a second set of call events including voice models substantially matching the second voice model.

22. The system according to claim 18, further comprising a voice model comparator stored in memory and executable by the processor to compare the first and second voice sample to a voice model for a customer associated with the customer account.

23. The system according to claim 22, wherein the voice model comparator is further executable to compare the first voice sample against voice models a blacklist that includes voice models of known fraudsters based on the first voice score, and to compare the second voice sample against voice models of known fraudsters based on the second voice score.

24. The system according to claim 22, further comprising:

a database module that stores agent voice models for comparison to segments of the first and second voice samples; and

a diarization module stored in memory and executable by the processor to receive agent voice models and remove segments of the first and second voice samples that include agent voices based the agent voice models.

25. The system according to claim 18, further comprising a database module stored in memory and executable by the processor to identify non-audio data associated with first and second call events that are linked to a fraudster.

26. The system according to claim 25, wherein the non-audio data includes any of an automatic number identification, caller identification information, an international mobile equipment identity number, a given name, the timestamp associated with the call event, keywords included in the call event, and combinations thereof.

27. The system according to claim 18, further comprising a diarization module that diarizes and removes segments of the first and second voice samples associated with agent voice models by:

segmenting the first and second voice samples;

detecting segments of voice samples corresponding to an agent in the first and second voice samples, using an agent voice model; and

removing the detected segments from the first and second voice samples.

28. The system according to claim 18, wherein each call event comprises a screening identification for associating the call event with the customer account.

29. A non-transitory computer readable storage media having a program embodied thereon, the program being executable by a processor to perform a method for disambiguating call data, the method comprising:

receiving notification of a fraud event associated with a customer account, the fraud event comprising a fraud time stamp;

extracting a first voice model and a second voice model from voice samples in call events obtained within a time frame that is temporally proximate the fraud time stamp;

determining for each call event within the time frame if the voice sample matches the first voice model or the second voice model; and

generating a chart that includes for each call event within the time frame a column for a time stamp and a column for an indication that the voice sample matches the first voice model or the second voice model.

30. The non-transitory computer readable storage media according to claim 29, wherein the chart is a spreadsheet.

31. The non-transitory computer readable storage media according to claim 29, wherein the chart includes for each call event a column for a score representing a match between the respective first and second voice samples and a voice model in a blacklist.

32. The non-transitory computer readable storage media according to claim 29, wherein the chart includes for each call event a column for a score representing a match between the respective first and second voice samples and a voice model in a whitelist.

33. A method for disambiguating call data, the method comprising:

receiving, via a fraud notification module, notification of a fraud event associated with a customer account, the fraud event comprising a time stamp;

selecting, using a call selection module, call events occurring within a predetermined time based on the fraud time stamp, each call event including a voice model;

comparing, using a voice model comparator, the included voice models to a customer voice model associated with the customer account; and

storing, in a fraudster database, included voice models that do not match the customer voice model.