Systems and methods for detecting illegitimate voice calls
The disclosed computer-implemented method for detecting illegitimate voice calls may include (1) identifying an incoming voice call, (2) processing the incoming voice call in real time by (a) segmenting the incoming voice call into progressively produced call segments and, (b) for each new segment as the progressively produced call segments are produced, (A) extracting a set of features from the new segment and (B) feeding, as input into a neural network, the set of features and an output from the neural network generated based on a preceding segment of the incoming voice call, thereby generating a new output representing the current likelihood that the incoming voice call is illegitimate, (3) determining that the likelihood that the incoming voice call is illegitimate is above a predetermined threshold, and (4) performing a security action during the incoming voice call. Various other methods, systems, and computer-readable media are also disclosed.
1. A computer-implemented method for detecting illegitimate voice calls, at least a portion of the method being performed by a computing device comprising at least one processor, the method comprising:
identifying, at a computing system that receives voice calls, an incoming voice call;
processing the incoming voice call at the computing system in real time by:
segmenting the incoming voice call into progressively produced call segments; and
for each new segment as the progressively produced call segments are produced:
extracting a set of features from the new segment; and
feeding, as input into a neural network, the set of features from the new segment and an output from the neural network generated based on a preceding segment of the incoming voice call, thereby generating a new output representing a current assessment of a likelihood that the incoming voice call is illegitimate;
determining that the likelihood that the incoming voice call is illegitimate is above a predetermined threshold based on output from the neural network; and
performing a security action on the incoming voice call during the incoming voice call upon determining that the likelihood that the incoming voice call is illegitimate is above a predetermined threshold.
2. The computer-implemented method of claim 1 , wherein the security action comprises at least one of:
alerting a user of the computing system during the incoming voice call about the likelihood that the incoming voice call is illegitimate;
muting the incoming voice call; and
terminating the incoming voice call.
3. The computer-implemented method of claim 1 , wherein the neural network was trained before the incoming voice call using a plurality of sample voice calls and a plurality of legitimacy classifications applied to the plurality of sample voice calls.
4. The computer-implemented method of claim 1 , wherein the set of features describes content of the incoming voice call.
5. The computer-implemented method of claim 1 , wherein:
extracting the set of features from the new segment comprises converting speech within the new segment to textual features; and
feeding, as input into the neural network, the set of features comprises providing the textual features as input into the neural network.
6. The computer-implemented method of claim 1 , wherein the set of features comprises at least one of:
at least characteristic of background noise observed during the new segment; and
at least one non-textual speech characteristic observed during the new segment.
7. The computer-implemented method of claim 1 , wherein the computing system comprises a mobile phone.
8. The computer-implemented method of claim 1 , further comprising:
receiving, from a user of the computing system who received the incoming voice call, a legitimacy classification of the incoming voice call; and
further training the neural network based on the legitimacy classification.
9. The computer-implemented method of claim 1 , wherein the neural network comprises a recurrent neural network.
10. The computer-implemented method of claim 1 , wherein processing the incoming voice call at the computing system is in response to determining that a source of the incoming voice call comprises a non-trusted source.
11. A system for detecting illegitimate voice calls, the system comprising:
an identification module, stored in memory, that identifies, at a computing system that receives voice calls, an incoming voice call;
a processing module, stored in memory, that processes the incoming voice call at the computing system in real time by:
segmenting the incoming voice call into progressively produced call segments; and
for each new segment as the progressively produced call segments are produced:
extracting a set of features from the new segment; and
feeding, as input into a neural network, the set of features from the new segment and an output from the neural network generated based on a preceding segment of the incoming voice call, thereby generating a new output representing a current assessment of a likelihood that the incoming voice call is illegitimate;
a determination module, stored in memory, that determines that the likelihood that the incoming voice call is illegitimate is above a predetermined threshold based on output from the neural network;
a performing module, stored in memory, that performs a security action on the incoming voice call during the incoming voice call upon determining that the likelihood that the incoming voice call is illegitimate is above a predetermined threshold; and
at least one physical processor configured to execute the identification module, the processing module, the determination module, and the performing module.
12. The system of claim 11 , wherein the security action comprises at least one of:
alerting a user of the computing system during the incoming voice call about the likelihood that the incoming voice call is illegitimate;
muting the incoming voice call; and
terminating the incoming voice call.
13. The system of claim 11 , wherein the neural network was trained before the incoming voice call using a plurality of sample voice calls and a plurality of legitimacy classifications applied to the plurality of sample voice calls.
14. The system of claim 11 , wherein the set of features describes content of the incoming voice call.
15. The system of claim 11 , wherein:
the processing module extracts the set of features from the new segment by converting speech within the new segment to textual features; and
the processing module feeds, as input into the neural network, the set of features by providing the textual features as input into the neural network.
16. The system of claim 11 , wherein the set of features comprises at least one of:
at least characteristic of background noise observed during the new segment; and
at least one non-textual speech characteristic observed during the new segment.
17. The system of claim 11 , wherein the computing system comprises a mobile phone.
18. The system of claim 11 , wherein the performing module further:
receives, from a user of the computing system who received the incoming voice call, a legitimacy classification of the incoming voice call; and
trains the neural network further based on the legitimacy classification.
19. The system of claim 11 , wherein the neural network comprises a recurrent neural network.
20. A non-transitory computer-readable medium comprising one or more computer-readable instructions that, when executed by at least one processor of a computing device, cause the computing device to:
identify, at a computing system that receives voice calls, an incoming voice call;
process the incoming voice call at the computing system in real time by:
segmenting the incoming voice call into progressively produced call segments; and
for each new segment as the progressively produced call segments are produced:
extracting a set of features from the new segment; and
feeding, as input into a neural network, the set of features from the new segment and an output from the neural network generated based on a preceding segment of the incoming voice call, thereby generating a new output representing a current assessment of a likelihood that the incoming voice call is illegitimate;
determine that the likelihood that the incoming voice call is illegitimate is above a predetermined threshold based on output from the neural network; and
perform a security action on the incoming voice call during the incoming voice call upon determining that the likelihood that the incoming voice call is illegitimate is above a predetermined threshold.