US 12,764,226 B1Granted

Fraudulent call detection in a contact center

AssigneeWells Fargo Bank, N.A.

InventorsRameshchandra B. Ketharaju, Anindya Sil

Life of this patent

Estimating the term…

Loading patent…

Abstract

A call center computing system (CCCS) identifies a user profile associated with information identifying a user in a call received at a call center. The CCCS accesses an audio segment of the call and applies a fraud detection model to the audio segment to determine one or more scores predicting a likelihood the call is fraudulent. Applying the fraud detection model includes accessing a cultural tone profile defining a first combination of features determined from a baseline voice sample associated with the user profile and generating prosodic features defining a second combination of features associated with a voice recorded in the audio segment. Applying the fraud detection model also includes determining the one or more scores by evaluating the generated prosodic features based on the cultural tone profile. The CCCS categorizes the call in a particular category selected from a set of categories and performs an action with the call.

Claims18

Dependent claims

A computer-implemented method in which a computing system performs operations comprising:

identifying a user profile associated with information identifying a user in a call received at a call center;
accessing, by the computing system, audio data comprising an audio segment of the call received at the call center;
applying a fraud detection model to the audio data to determine one or more scores predicting a likelihood that the call is fraudulent, wherein applying the fraud detection model to the audio data comprises:
accessing a cultural tone profile associated with the user profile, wherein the cultural tone profile defines a first combination of features determined from a baseline voice sample associated with the user profile, wherein the cultural tone profile is one of a plurality of cultural tone profiles generated by;
accessing, for each of a set of user profiles including the user profile, a respective voice sample, wherein the voice sample accessed for the user profile comprises the baseline voice sample;
clustering, based on a set of features extracted from the accessed voice samples, the accessed voice samples into a set of clusters;
and
based on the clustering:
generating the plurality of cultural tone profiles by defining, for each of the set of clusters, a cultural tone profile having a specific combination of values associated with the set of features;
and
assigning each user profile to a respective cultural tone profile of the plurality of cultural tone profiles;
generating prosodic features defining a second combination of features associated with a voice recorded in the audio segment of the call;
and
determining the one or more scores by evaluating the generated prosodic features based on the cultural tone profile;
based on the one or more scores, categorizing the call in a particular category selected from a predefined set of categories;
and
based on the selected category for the call, performing an action with regard to the call.

14 more elements

A non-transitory computer-readable medium having program code that is stored thereon, the program code executable by one or more processing devices for performing operations comprising:

identifying a user profile associated with information identifying a user in a call received at a call center;
accessing, by the one or more processing devices, audio data comprising an audio segment of the call received at the call center;
applying a fraud detection model to the audio data to determine one or more scores predicting a likelihood that the call is fraudulent, wherein applying the fraud detection model to the audio data comprises:
accessing a cultural tone profile associated with the user profile, wherein the cultural tone profile defines a first combination of features determined from a baseline voice sample associated with the user profile, wherein the cultural tone profile is one of a plurality of cultural tone profiles generated by;
accessing, for each of a set of user profiles including the user profile, a respective voice sample, wherein the voice sample accessed for the user profile comprises the baseline voice sample;
clustering, based on a set of features extracted from the accessed voice samples, the accessed voice samples into a set of clusters;
and
based on the clustering:
generating the plurality of cultural tone profiles by defining, for each of the set of clusters, a cultural tone profile having a specific combination of values associated with the set of features;
and
assigning each user profile to a respective cultural tone profile of the plurality of cultural tone profiles;
generating prosodic features defining a second combination of features associated with a voice recorded in the audio segment of the call;
and
determining the one or more scores by evaluating the generated prosodic features based on the cultural tone profile;
based on the one or more scores, categorizing the call in a particular category selected from a predefined set of categories;
and
based on the selected category for the call, performing an action with regard to the call.

14 more elements

A system, comprising:

a non-transitory computer readable medium comprising computer-executable program instructions that, when executed by a processor, cause the system to perform operations comprising:
identifying a user profile associated with information identifying a user in a call received at a call center;
accessing, by the system, audio data comprising an audio segment of the call received at the call center;
applying a fraud detection model to the audio data to determine one or more scores predicting a likelihood that the call is fraudulent, wherein applying the fraud detection model to the audio data comprises:
accessing a cultural tone profile associated with the user profile, wherein the cultural tone profile defines a first combination of features determined from a baseline voice sample associated with the user profile, wherein the cultural tone profile is one of a plurality of cultural tone profiles generated by;
accessing, for each of a set of user profiles including the user profile, a respective voice sample, wherein the voice sample accessed for the user profile comprises the baseline voice sample;
clustering, based on a set of features extracted from the accessed voice samples, the accessed voice samples into a set of clusters;
and
based on the clustering:
generating the plurality of cultural tone profiles by defining, for each of the set of clusters, a cultural tone profile having a specific combination of values associated with the set of features;
and
assigning each user profile to a respective cultural tone profile of the plurality of cultural tone profiles
generating prosodic features defining a second combination of features associated with a voice recorded in the audio segment of the call;
and
determining the one or more scores by evaluating the generated prosodic features based on the cultural tone profile;
based on the one or more scores, categorizing the call in a particular category selected from a predefined set of categories;
and
based on the selected category for the call, performing an action with regard to the call.

15 more elements

Dependencies are inferred from the claim text.