New PADSE Project Aims to Combat Audio Deepfakes

The PADSE (Person-Centric Audio- and Speech-based Deepfake and Shallowfake Detection) research project has been launched by Fraunhofer IDMT, Deutsche Welle, and Bauhaus-Universität Weimar to develop advanced methods for detecting audio deepfakes. As audio deepfakes become increasingly realistic, existing detection methods are struggling to keep pace. The project introduces person-centric reference profiles that account for individual voice characteristics, speaking styles, and emotional expressions. This innovative approach aims to enhance the reliability of detecting manipulated speech content. The project is application-oriented, focusing on real-world evaluations of the developed methods. The consortium combines expertise in audio forensics, text analysis, and journalistic practices to address privacy and security challenges from the outset. The technologies developed are intended to assist media organizations, companies, and public authorities in identifying synthetic speech content more effectively.

Key Points: • PADSE project launched to enhance audio deepfake detection methods. • Person-centric reference profiles will improve detection accuracy. • The project combines expertise from academia and media organizations.