Virtual Booth

The rapid advancement of artificial intelligence has enabled highly realistic speech synthesis, voice cloning, and deepfake audio generation technologies. While these innovations enhance human computer interaction, they are increasingly exploited for malicious activities such as phone scams, voice phishing, impersonation fraud, and robocall attacks. As voice-based communication plays a vital role in online transactions and sensitive interactions, Traditional rule-based detection methods are insufficient to fight sophisticated AI-generated voices. This project proposes an AI-powered deepfake audio scam detection system capable of identifying fraudulent and manipulated speech in both real-time and offline scenarios. The system employs a Fast CNN-LSTM hybrid architecture, where Convolutional Neural Networks extract discriminative spectral features from MFCCs and Mel-spectrograms, and Long Short-Term Memory networks model temporal speech patterns to detect spoofing inconsistencies. Experimental results demonstrate high detection accuracy, making the system suitable for deployment in financial institutions, telecommunication networks, and cybersecurity platforms to enhance protection against voice-based fraud.