Web Crawler for detecting audio DeepFakes

Wrocław University of Science and Technology 2024 ZPI Day Submission17 Authors

Keywords: EarDefender, audio DeepFake, DeepFake detection, audio scraping, wav2vec 2.0

Abstract: The proliferation of Audio DeepFakes has enabled the creation of highly convincing yet deceptive audio, leading to critical threats such as identity theft and the spread of misinformation. As DeepFake generation tools become increasingly accessible, coupled with the ease of sharing manipulated files across social media and other platforms, the global impact of these threats is escalating. To address this challenge, we present EarDefender, a web-based application designed to detect and combat audio DeepFakes. EarDefender uses cutting-edge AI models, SSL-wav2vec, and MesoNet, to ensure robust detection of manipulated audio. The platform scans the chosen Internet segment, verifies audio authenticity, and generates detailed summary reports, enabling users to identify and mitigate DeepFake risks easily, regardless of technical expertise. Through rigorous evaluation, the application has demonstrated high performance in delivering fast and reliable detection. EarDefender marks a significant advancement in safeguarding online spaces from the dangers of audio manipulation.

Submission Number: 17