Deep fake speech and automatic speaker verification systems in the African setting

Journal Title
Journal ISSN
Volume Title
Automatic Speaker Verification (ASV) systems authenticate anyone who interacts with a digital system using speech in a seamless manner. ASV systems are widely utilised in the banking industry to validate customer identities. Deep neural network (DNN)- based voice synthesis systems, on the other hand, have enabled the cloning of the human voice in the form of deepfake audio, which can fool most humans and ASV systems. If these deepfake audios are used maliciously, they jeopardise people's identities and impede the security of ASV systems. This study documents efforts and findings from an exhaustive experimental study on the impact of deepfake audio generated on African accents on ASV systems. We found that deepfake audio generated on African accents is less likely to fool modern ASV systems. These findings highlight the importance of deepfake audio systems that can simulate convincing African accents to ensure that current technologies are used to tackle modern challenges in Africa.
Undergraduate thesis submitted to the Department of Computer Science and Information Systems, Ashesi University, in partial fulfillment of Bachelor of Science degree in Computer Science, May 2022