Analyzing Exposure in Generative Adversarial Networks: Advancing Security Against AI-Synthesized Voice Threats
Geda Wakene, Zizhao Chen, W. Eric Wong, Chih-Wei Hsu · 2025
As AI proceeds to continue its inevitable growth, the blossoming of deep fakes and AI-synthesized fake voices are beginning to pose significant challenges to the integrity of both audio and visual content on digital platforms. As technologies such as generative adversarial networks (GANs) become increasingly sophisticated, the demand to identify and mitigate the vulnerabilities within them becomes eminent. This study explores novel methodologies to uncover these vulnerabilities that are present in these fake voices and deep fakes. Through using existing techniques, such as a deep neural network (DNN), this research identifies potential weaknesses in the models underlying these technologies. With the knowledge of these weaknesses, the aim is to enhance the robustness of systems that may be prone to manipulation and deception by giving them the ability to both detect and exploit the vulnerabilities in AI-synthesized content to expose the fake from the real. Ultimately, these findings are for the purpose of contributing towards the ongoing efforts to safeguard the integrity and authenticity of multimedia content in the digital age.