Srivarshini I, Pillai Kirthika Kannan, Tamilselvi P, Swathi. G, Ranitha R
In this paper, the authors introduce a secure voice transaction system that uses deepfake technology to increase the validity of authentication in online financial transactions. Passwords and OTPs, which are commonly used as traditional authentication tools, are under increased cyber threat, such as voice spoofing attacks and AI-generated impersonation attacks. To solve these issues, the proposed system involves a two-layered security system that is based on a combination of biometric voice identification and deep learning-based fake voice recognition. In the verification of identity, Mel-Frequency Cepstral Coefficients (MFCC) are computed on the voice inputs of the user and the comparison between them is carried out by the use of cosine similarity. At the same time, a Convolutional Neural Network (CNN) is used to analyze Mel-spectrogram representations of the audio to identify it as an original or an artificial one. Approving a transaction requires fulfillment of background checks, as well as authenticity and deep fake checks. The system is installed on flask-based backend and voice capture over the browser. The experimental findings show better strength and resiliency to the contemporary voice-based fraud attacks.