Method and apparatus for fraud detection

• US 20060285665A1
• Filed: 05/27/2005
• Published: 12/21/2006
• Est. Priority Date: 05/27/2005
• Status: Active Grant

• Current Assignee: Nice Ltd
• Original Assignee: Nice Systems Limited (Nice Ltd)
• Inventors: Wasserblat, Moshe, Pereg, Oren, Kor, Ilan, Eilam, Barak
• Granted Patent: US 7,386,105 B2
• Time in Patent Office: Days
• US Class Current: 379/114.140
• CPC Class Codes: G06Q 20/4016 involving fraud or risk level assessment in transaction processing
• CPC Class Codes: G10L 17/26 Recognition of special voice characteristics, e.g. for use in lie detectors; Recognition of animal voices
• 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 3/36 Statistical metering, e.g. recording occasions when traffic exceeds capacity of trunks
• CPC Class Codes: H04M 3/385 using speech signals

Abstract

An apparatus and method for detecting a fraud or fraud attempt in a captured interaction. The method comprising a selection step in which interactions suspected as capturing fraud attempts are selected for further analysis, and assigned a first fraud probability, and a fraud detection step in which the voice is scored against one or more voice prints, of the same alleged customer or of known fraudsters. The first fraud or fraud attempt probability is combined with the result of the scoring of the fraud detection step, to generate a total fraud or fraud attempt probability. If the total fraud or fraud attempt probability exceeds a threshold, a notification is issued. The selection, scoring and combination thereof are performed using user-defined rules and thresholds.

Claims

1. A fraud detection method for generating a first fraud or fraud attempt probability, within an at least one captured or recorded interaction, the method comprising the step of scoring an at least one voice belonging to an at least one tested speaker in the at least one captured or recorded interaction against an at least one voice print within an at least one entry in a voice print collection, the scoring step generating a probability that the at least one voice in the captured or recorded interaction belongs to an at least one second speaker associated with the at least one voice print, said probability represents the probability that the at least one captured or recorded interaction is fraudulent.

2. The method of claim 1 wherein the probability generated at the scoring step represents the similarity between the voice of the tested speaker and the voice of the at least one second speaker.

3. The method of claim 1 wherein the at least one second speaker is at least one of the group consisting of:

a fraudster;

a legitimate speaker;

or allegedly the tested speaker.

4. The method of claim 1 further comprising a fraud probability generation step, for enhancing the first fraud or fraud attempt probability.

5. The method of claim 4 wherein the fraud probability generation step employs a rule engine activating an at least one rule or using an at least one threshold set by a user of the method.

6. The method of claim 5 wherein said at least one rule or at least one threshold involves data associated with:

the captured or recorded interaction;

a speaker thereof;

an at least one other interaction of the speaker thereof;

the at least one voice print;

or the at least one second speaker.

7. The method of claim 5 wherein the at least one rule or the at least one threshold is set dynamically.

8. The method of claim 1 further comprising:

a second scoring step for determining a second scoring result by scoring the at least one voice against a background model; and

a normalizing step for normalizing the first scoring result with the second scoring result.

9. The method of claim 1 wherein the scoring step comprises scoring the tested voice against at least two voice prints, thus obtaining at least two score results.

10. The method of claim 9 further comprising a score-test normalization step, for normalizing the at least two score results.

11. The method of claim 1 wherein the at least one voice print or data associated with the at least one voice print or with the second speaker is stored in a collection.

12. The method of claim 1 further comprising:

a retrieving step for retrieving an at least one second interaction; and

a construction step for constructing the at least one voice print from an at least one voice participating in the at least one second interaction.

13. The method of claim 1 further comprising a step of retrieving an at least one voice print or retrieving data associated with the at least one voice print.

14. The method of claim 13 further comprising a filtering step for filtering a collection to retrieve the at least one voice print such that data associated with the at least one voice print is related to data associated with the at least one recorded or captured interaction.

15. The method of claim 14 wherein the filtering is based on an at least one of the group consisting of:

dynamic data associated with the at least one voice recorded or captured interaction, dynamic data associated with the at least one voice print, predetermined data associated with the at least one voice recorded or captured interaction, predetermined data associated with the at least one voice print, area code of calling number, gender of speaker, age pf speaker, language spoken.

16. The method of claim 1 wherein the steps are performed in real-time.

17. The method according to claim 1 further comprising a pre processing step for pre processing the captured or recorded interaction.

18. The method of claim 17 wherein the pre processing step comprises at least one of the group consisting of:

compressing;

decompressing;

segmenting;

separating speakers;

identifying the agent side; and

receiving data related to the at least one captured or recorded interaction.

19. The method of claim 18 wherein the additional data comprises computer telephony integration information.

20. The method according to claim 12 further comprising a pre processing step for pre processing the at least one second interaction.

21. The method of claim 1 further comprising a selection step for selecting an at least one selected interaction to be transferred to the scoring step, and associating the at least one selected interaction with a second fraud or fraud attempt probability.

22. The method of claim 21 wherein the selection step comprises at least one of the group consisting of:

applying an at least one rule by a rule engine;

auditing interactions;

filtering interactions using an at least one filtering engine.

23. The method of claim 22 wherein the at least one filtering engine comprises one or more of the group consisting of:

emotion detection, word spotting, speech to text, interaction analysis, and data risk analysis.

24. The method of claim 21 wherein the second fraud or fraud attempt probability represents a risk level associated with the at least one selected interaction.

25. The method of claim 21 wherein the selection step employs an at least one rule or an at least one threshold set by a user of the method.

26. The method of claim 25 wherein said rule or threshold involves data associated with at least one of the group consisting of:

the captured or recorded interaction;

a speaker thereof;

other interactions of the speaker thereof;

the at least one voice print or the at least one second speaker.

27. The method of claim 25 wherein the at least one rule or the at least one threshold is set dynamically.

28. The method of claim 21 further comprising a weighting step for generating a combined fraud attempt probability from the first and the second fraud attempt probabilities.

29. The method according to claim 1 further comprising an alert-generation step for generating an alert if said first fraud or fraud attempt probability exceeds a predetermined threshold.

30. The method of claim 29 wherein the alert is sent to an agent or another contact associated with the captured or recorded interaction.

31. The method according to claim 21 further comprising an alert-generation step for generating an alert if said first fraud or fraud attempt probability or said second fraud or fraud attempt probability or a combination thereof exceeds a predetermined threshold.

32. The method according to claim 1 further comprising a reporting step for generating or updating a report comprising details of the at least one captured or recorded interaction if said first fraud or fraud attempt probability exceeds a predetermined threshold.

33. The method of claim 32 further comprising an auditing step for auditing the report.

34. The method of claim 33 wherein auditing the report comprises at least one of the group consisting of:

listening to calls;

reviewing interactions;

reviewing history of at least one speaker participating in an at least one interaction;

listening to previous interactions of speakers participating in an at least one interaction appearing in the report, or filling forms.

35. The method according to claim 21 further comprising a reporting step for generating or updating a report comprising details of the at least one captured or recorded interaction if said first fraud or fraud attempt probability or said second fraud or fraud attempt probability or a combination thereof exceeds a predetermined threshold.

36. The method of claim 1 further comprising a step of interaction capturing or logging.

37. The method of claim 1 further comprising a step of a-posteriori analysis for at least one of the group consisting of:

adapting an at least one parameter or an at least one rule or an at least one threshold used by the method, trend analysis, pattern of behavior recognition, selecting an at least one filtering engine.

38. A method for constructing a voice print collection, the method comprising the steps of:

receiving at least one suspect voice sample;

extracting characteristic features from the at least one voice sample, thus generating an at least one sequence of feature vectors;

constructing at least one voice print from the at least one sequence of feature vectors; and

storing the at least one voice print in a collection.

39. The method of claim 38 further comprising a storage step for storing data in the collection, said data associated with the at least one voice print, or with the at least one voice sample, or with an at least one speaker participating in the voice sample.

40. The method of claim 38 further comprising a normalization step for generating score normalization parameters to be used when scoring an at least one voice sample against an at least one voice print from the collection.

41. A fraud detection apparatus for generating a fraud or fraud attempt probability, within an at least one captured or recorded interaction, the apparatus comprising:

a voice print collection comprising at least one entry, each entry comprising a voice print or data associated with the voice print, said voice print is constructed from a voice known or suspected as belonging to a fraudster; and

a scoring component for scoring an at least one voice in the at least one captured or recorded interaction against the voice print within the at least one entry, the component generating a probability that the at least one voice in the at least one captured or recorded interaction belongs to an at least one person associated with the voice print, said probability represents the probability that the at least one captured or recorded interaction is fraudulent.