Study on the Effect of Face Masks on Forensic Speaker Recognition

Abstract

The COVID-19 pandemic has led to a dramatic increase in the use of face masks. Face masks can affect both the acoustic properties of the signal and the speech patterns and have undesirable effects on automatic speech recognition systems as well as on forensic speaker recognition and identification systems. This is because the masks introduce both intrinsic and extrinsic variability into the audio signals. Moreover, their filtering effect varies depending on the type of mask used. In this paper we explore the impact of the use of different masks on the performance of an automatic speaker recognition system based on Mel Frequency Cepstral Coefficients to characterise the voices and on Support Vector Machines to perform the classification task. The results show that masks slightly affect the classification results. The effects vary depending on the type of mask used, but not as expected, as the results with FPP2 masks are better than those with surgical masks. An increase in speech intensity has been found with the FPP2 mask, which is related to the increased vocal effort made to counteract the effects of hearing loss.