Audio-deepfake: generation methods, legitimate applications and the potential for misuse

Audio deepfake has emerged as a powerful tool with the potential to revolutionise various fields, including virtual assistants, gaming, entertainment, education, audiobook production, and accessibility solutions. However, it also presents significant security issues and ethical concerns. The primary challenge posed by this new technology comes from the realistic and convincing nature of the generated audio clips, which are capable of fooling both humans and conventional security measures. Additionally, the growing accessibility of open-source deepfake tools, combined with rising computational capabilities, makes it easier for malicious actors to produce and spread audio deepfakes. In this paper, we present a detailed review of current audio deepfake generation methods, considering their ethical applications as well as their potential for abuse. We also identify key directions for future research to advance this domain.