1 paper
Vojtěch Staněk, Michal Frič, Kamil Malinka +1
Conventional deepfake speech detectors primarily target synthesis artifacts and often degrade under unseen attacks or recording conditions. Human speech, however, contains physiolo…