Unmasking Synthetic Speech: A Novel Deep Learning Model for Deepfake Audio Detection
Rejin Paul N R, J. Mary Hanna Priyadharshini, Arun Manicka Raja M, Shakthi Priya S · 2025
The deep fake audio generation advances have caused difficulty in differentiating between human and AI-generated speech. Insignificant threats due to misinformation and cybersecurity risks first and foremost. We further tackle the above problems and devise a deep learning method for deep fake audio detection presented in this paper. Our approach is based on CNNs and utilizes new improved versions of feature extraction that can outperform the generalization by regularization. We evaluate our model on different authentic and AI-based speech datasets to establish its broad compatibility in the context of changing deep fakes. Experiments indicate that our method achieves better performance than the previous state of the arts on accuracy and robustness as well reduce false-positive rates, thus improving detection quality. Our model gets an accuracy of around 92.5% we have for deep fake audio detection as a high-quality solution.