| CPC H04M 3/42085 (2013.01) [G10L 15/02 (2013.01); G10L 15/26 (2013.01)] | 18 Claims |
|
1. A system for identification of a fraudulent voice call in real-time, the system comprising:
a first device communicatively coupled through a network to a second device, the first device configured to:
receive a voice call originating from the second device;
determine an absence of details associated with the second device in the first device; and
generate a call forward request based on the absence of details;
a server communicatively coupled to the first device and the second device through the network, the server configured to:
receive the call forward request from the first device via the network; and in response to the call forward request:
divide the voice call from the second device to the first device into a plurality of audio signals in real time during the voice call;
analyze, using at least two artificial intelligence models, one or more audio signals originated from the second device of the plurality of audio signals to identify whether an audio originating from the second device is a cloned voice or a human voice, wherein analyzing the one or more audio signals to identify whether the audio is a cloned voice or a human voice includes:
determining, using a first artificial intelligence model of the at least two artificial intelligence models, a first confidence score indicative of whether the audio is a cloned voice or a human voice, wherein the first artificial intelligence model analyzes frequency components of the one or more audio signals originated from the second device during the voice call,
determining, using a second artificial intelligence model of the at least two artificial intelligence model, a second confidence score indicative of whether the audio is a cloned voice or a human voice, wherein the second artificial intelligence model is trained on datasets including a plurality of sample cloned voices and a plurality of sample human voices to identify the audio as a cloned voice or a human voice,
averaging the first confidence score and the second confidence score to determine an average confidence score, and
identifying the audio as the cloned voice when the average confidence score is less than a predefined threshold value;
analyze, upon reception of each audio signal of the plurality of audio signals, the audio signal along with one or more preceding audio signals to continuously determine a risk score associated with an identification of a fraudulent activity during the voice call:
identify the voice call as the fraudulent voice call upon identifying the audio originating from the second device as a cloned voice or when the risk score associated with the identification of the fraudulent activity during the voice call is greater than a predetermined risk value; and
trigger an alert in the first device during the voice call to indicate that the voice call from the second device is a fraudulent voice call in response to the identifying.
|