Bridging Music and Media: AI-Driven and Human-based Solutions for Detecting Audio Deepfakes in an Era of Misinformation

Abstract

Artificial intelligence (AI) has become a vital part of the 21st-century digital world, serving as vital tools in enhancing human creativity and daily tasks. While it contributes to improving human lives, the rise of AI-generated audio deepfakes poses unprecedented challenges to both music and mass communication. One of the major challenges posed by AI in this regard is its contribution to the proliferation of misinformation through synthetic voices, manipulated interviews, and unreal musical performances. This paper explored AI-driven solutions for detecting deepfakes and examined human-based strategies in the same light, given the constant advancement of generative AI models and their ability to evade detection. The study employed a mixed-method approach. First, an experimental approach was used in creating deepfake music, which was later subjected to scrutiny by several AI-powered detectors. Secondly, a qualitative approach was used to source experts’ recommendations on human-based strategies for combating deepfakes through available online interviews. Data was analyzed using content analysis. Findings revealed that existing AI-driven detection tools often misclassify or fail to identify AI-generated audio due to the continuous advancement in the development and updating of generative AI models. Human-based strategies such as critical listening and attention can help to identify robotic timbres, unnatural phrasing, the lack of breath, slips between frames that make the face jiggle, and linguistic imbalance in tonal languages. Verification through official channels and record labels is also a key human-centered approach in detecting deepfakes and should complement the use of AI-based detectors. Future research should prioritize public perception of AI audio and the efficacy of media literacy across demographics, the cultural impact on music authenticity, the development of global regulatory frameworks, and the differential interpretation of multimodal versus audio-only deepfakes.