Explainable Deepfake and Spoofing Detection: An Attack Analysis Using SHapley Additive exPlanations
Wanying Ge, Massimiliano Todisco, Nicholas Evans
Abstract
Despite several years of research in deepfake and spoofing detection for\nautomatic speaker verification, little is known about the artefacts that\nclassifiers use to distinguish between bona fide and spoofed utterances. An\nunderstanding of these is crucial to the design of trustworthy, explainable\nsolutions. In this paper we report an extension of our previous work to better\nunderstand classifier behaviour to the use of SHapley Additive exPlanations\n(SHAP) to attack analysis. Our goal is to identify the artefacts that\ncharacterise utterances generated by different attacks algorithms. Using a pair\nof classifiers which operate either upon raw waveforms or magnitude\nspectrograms, we show that visualisations of SHAP results can be used to\nidentify attack-specific artefacts and the differences and consistencies\nbetween synthetic speech and converted voice spoofing attacks.\n