Method for training a spoofing detection model using biometric clustering

1. A system comprising:

a processor and a computer readable medium operably coupled thereto, the computer readable medium comprising a plurality of instructions stored in association therewith that are accessible to, and executable by, the processor, to perform operations which comprise:

receiving a plurality of customer call interactions;

classifying, by a spoofing engine, each of the plurality of customer call interactions as a spoofed call or a non-spoofed call using a spoofing detection model;

generating, by a proactive fraud exposure (PFE) engine, a voiceprint for each of the plurality of customer call interactions;

comparing, by the PFE engine, the generated voiceprints;

grouping, by the PFE engine, the generated voiceprints into one or more clusters based on the comparing, wherein each cluster represents a single speaker;

locating a cluster containing a spoofed call and a non-spoofed call, thereby indicating that the non-spoofed call was misclassified by the spoofing detection model; and

updating the spoofing detection model with the non-spoofed call.

2. The system of claim 1 , wherein:

comparing the generated voiceprints comprises comparing pairs of generated voiceprints and calculating similarity scores for each pair, and

the operations further comprise establishing a threshold similarity score.

3. The system of claim 2 , wherein each pair of generated voiceprints is grouped into a cluster when the similarity score for the pair exceeds the threshold similarity score.

4. The system of claim 1 , wherein the operations further comprise calculating a confidence score for each cluster by comparing all the generated voiceprints in each cluster against each other.

5. The system of claim 4 , wherein the operations further comprise ranking the plurality of clusters from high to low based on the confidence score and using the highest-ranked clusters to locate the cluster containing a spoofed call and a non-spoofed call.

6. The system of claim 1 , wherein the operations further comprise:

classifying each of the plurality of customer call interactions as a spoofed call or a non-spoofed call using the updated spoofing detection model;

detecting a cluster containing a spoofed call and a non-spoofed call, thereby indicating that the non-spoofed call in the detected cluster was misclassified by the updated spoofing detection model; and

re-updating the updated spoofing detection model with the non-spoofed call in the detected cluster.

7. The system of claim 1 , wherein classifying each of the plurality of customer call interactions as a spoofed call or a non-spoofed call comprises:

extracting speech features from an audio stream of each customer call interaction; and

predicting whether the extracted speech features indicate a spoofed call or a non-spoofed call.

8. The system of claim 1 , wherein the operations further comprise transmitting the non-spoofed call to a security officer for confirmation that the non-spoofed call was misclassified.

9. The system of claim 1 , wherein the operations further comprise:

receiving a plurality of additional customer call interactions; and

updating the spoofing detection model with the additional customer call interactions.

10. A method of training a spoofing detection model, which comprises:

receiving a plurality of customer call interactions;

classifying, by a spoofing engine, each of the plurality of customer call interactions as a spoofed call or a non-spoofed call using a spoofing detection model;

generating, by a proactive fraud exposure (PFE) engine, a voiceprint for each of the plurality of customer call interactions;

comparing, by the PFE engine, the generated voiceprints;

grouping, by the PFE engine, the generated voiceprints into one or more clusters based on the comparing, wherein each cluster represents a single speaker;

locating a cluster containing a spoofed call and a non-spoofed call, thereby indicating that the non-spoofed call was misclassified by the spoofing detection model; and

updating the spoofing detection model with the non-spoofed call.

11. The method of claim 10 , further comprising calculating a confidence score for each cluster by comparing all the generated voiceprints in each cluster against each other.

12. The method of claim 11 , further comprising ranking the plurality of clusters from high to low based on the confidence score and using the highest-ranked clusters to locate the cluster containing a spoofed call and a non-spoofed call.

13. The method of claim 10 , further comprising:

classifying each of the plurality of customer call interactions as a spoofed call or a non-spoofed call using the updated spoofing detection model;

detecting a cluster containing a spoofed call and a non-spoofed call, thereby indicating that the non-spoofed call in the detected cluster was misclassified by the updated spoofing detection model; and

re-updating the updated spoofing detection model with the non-spoofed call in the detected cluster.

14. The method of claim 10 , wherein classifying each of the plurality of customer call interactions as a spoofed call or a non-spoofed call comprises:

extracting speech features from an audio stream of each customer call interaction; and

predicting whether the extracted speech features indicate a spoofed call or a non-spoofed call.

15. The method of claim 10 , further comprising:

receiving a plurality of additional customer call interactions; and

updating the spoofing detection model with the additional customer call interactions.

16. A non-transitory computer-readable medium having stored thereon computer-readable instructions executable by a processor to perform operations which comprise:

receiving a plurality of customer call interactions;

classifying, by a spoofing engine, each of the plurality of customer call interactions as a spoofed call or a non-spoofed call using a spoofing detection model;

generating, by a proactive fraud exposure (PFE) engine, a voiceprint for each of the plurality of customer call interactions;

comparing, by the PFE engine, the generated voiceprints;

grouping, by the PFE engine, the generated voiceprints into one or more clusters based on the comparing, wherein each cluster represents a single speaker;

locating a cluster containing a spoofed call and a non-spoofed call, thereby indicating that the non-spoofed call was misclassified by the spoofing detection model; and

updating the spoofing detection model with the non-spoofed call.

17. The non-transitory computer-readable medium of claim 16 , wherein the operations further comprise:

calculating a confidence score for each cluster by comparing all the generated voiceprints in each cluster against each other;

ranking the plurality of clusters from high to low based on the confidence score; and

using the highest-ranked clusters to locate the cluster containing a spoofed call and a non-spoofed call.

18. The non-transitory computer-readable medium of claim 16 , wherein the operations further comprise:

classifying each of the plurality of customer call interactions as a spoofed call or a non-spoofed call using the updated spoofing detection model;

detecting a cluster containing a spoofed call and a non-spoofed call, thereby indicating that the non-spoofed call in the detected cluster was misclassified by the updated spoofing detection model; and

re-updating the updated spoofing detection model with the non-spoofed call in the detected cluster.

19. The non-transitory computer-readable medium of claim 16 , wherein classifying each of the plurality of customer call interactions as a spoofed call or a non-spoofed call comprises:

extracting speech features from an audio stream of each customer call interaction; and

predicting whether the extracted speech features indicate a spoofed call or a non-spoofed call.

20. The non-transitory computer-readable medium of claim 16 , wherein the operations further comprise:

receiving a plurality of additional customer call interactions; and

updating the spoofing detection model with the additional customer call interactions.