A place for linguistics in combating disinformation by audio deepfakes

Abstract

Deepfakes and their potential as vehicles for deception and disinformation pose some of the most pressing new challenges of our time. This paper includes a review of existing literature on automatic detection and human discernment of audio deepfakes in particular and identifies three major areas that lend themselves to further study. These areas include (1) the interaction between a voice’s gender and our ability to accurately identify it as fake or real, (2) how an understanding of language variation is relevant to the issue of deepfake detection and discernment, and (3) the potential for socio-variationist and perceptual linguistic principles to be incorporated into educational programs that improve listeners’ ability to differentiate between genuine speech and deepfakes. Findings support a crucial role for linguists to play in addressing the broader societal challenges surrounding misinformation and disinformation.