With the rise of voice assistants and voice-based logins, securing these systems against fake audio like recordings or AI-generated speech is crucial. This paper proposes a two-step voice authentication system for enhanced security. First, it uses WavLM with a lightweight classifier to detect whether a voice is genuine or fake. If the voice is real, it then verifies the speaker’s identity using ECAPA-TDNN, which generates a unique voiceprint. The system was trained and evaluated on the ASVspoof 2019 dataset, incorporating pre-processing steps like noise reduction and volume normalization. The experimental results, evaluated using EER, demonstrate high accuracy in both detecting spoofed voices and verifying the speaker identity. This two-stage pipeline proves to be not only effective but also scalable and well-suited for practical deployment in real-world voice authentication systems.