Deepfake Audio Detection on YouTube: Influences of Source Platform, Speaker, Language, and Age
Keywords
Loading...
Authors
Issue Date
2024-06-24
Language
en
Document type
Journal Title
Journal ISSN
Volume Title
Publisher
Title
ISSN
Volume
Issue
Startpage
Endpage
DOI
Abstract
Previous studies have highlighted deepfake as one of the most dangerous forms of
deception. However, studies investigating the human ability to detect deepfake is scarce,
especially in music audios. Therefore, this experiment investigates the human ability to detect deepfake audios on YouTube. Twelve audios, specifically songs, were displayed to n = 68 individuals and they were asked if they thought the audio was real. The experiment was
carried out in English to determine whether native language affects the detection accuracy. In
addition, it was tested whether age, presence of source and speaker familiarity affect detection accuracy of real and fake audios, as these variables are important in the deepfake field. The experiment was carried out in an online questionnaire, as a within-subject design. The whole sample showed an effect of speaker familiarity and age, but only for the fake audios. The more voices were recognized, the higher the detection accuracy. Regarding age, the detection accuracy decreased as the age increased. No effect was found for native language and presence of source.
Description
Citation
Supervisor
Faculty
Faculteit der Letteren