Recent algorithms for artificial voice synthesis have posed new fascinating opportunities, together with significant challenges across multiple domains (biometrics, fraud detection, authentication) and strong impacts on communication, security, and trust in the digital age. 

This thesis aims to explore the landscape of audio deepfakes, examining their technical foundations and developing new strategies to detect them for aduio sequence verification and the misuse mitigation. More specifically, candidates are going to focus on incremental and continual learning strategies, which allow adapting pre-trained models to new data characterized by evolving acquisition setups and enhanced synthesis approach. The candidate needs to possess some previous knowledge of deep learning and Python programming.