The rise of synthetic audio forgeries has created serious threats to digital communication, including impersonation, misinformation, and fraud. Conventional methods relying on handcrafted features such as MFCCs which are ineffective against advanced neural synthesis techniques like Text-to-Speech (TTS) and Voice Conversion (VC). This survey reviews recent advancements in audio deepfake detection using deep learning models, including CNNs, attention-based architectures, and hybrid frameworks. While detection performance has improved substantially, challenges remain in achieving strong generalization, robustness to adversarial attacks, and model interpretability. The paper also outlines key datasets, open research issues, and future directions for building more reliable audio forgery detection systems.