Ensuring transaction security is essential in the rapidly shifting world of digital payments. A deep learning-based voice authentication system called Safe-Voice-UPI attempts to fix the flaws in the Unified Payments Interface (UPI) framework of India, such as fraud, phishing, and illegal access. The project incorporates biometric voice recognition into the UPI system by utilizing distinctive vocal traits like pitch and tone, offering a reliable and convenient substitute for more conventional security measures like PINs. By providing improved fraud prevention, real-time monitoring, and user authentication, Safe-Voice-UPI has the potential to completely transform digital transaction security. This paper describes the development, integration, and testing methodologies of this system.