Deepfake audio detection with vision transformer based method

In addition to easy access to audio on the Internet, developments in deep learning methods have made it possible to produce deep fake audio. Deep fake audio spoofing is carried out with the aim of producing audio files in the content by cloning the voice of the speaker that is planned to be changed as if he said something he did not say. This forgery method, created using artificial intelligence approaches, poses a great threat, especially to speaker verification systems. This study proposes a new method to detect whether the audio is spoofed or original. Spectrogram images are obtained with the Constant-Q Transform approach then they are classified with the vision transformer network. The system is trained with a public dataset named ASVSpoof 2019 and the comparative performance analysis is done on this dataset.