Classification of Pitch Disguise Level with Artificial Neural Networks

Abstract

In audio forensics, voice disguise raises many serious challenges in finding the criminal suspects. Most of the criminals usually disguise their voice just before and/or after committing a crime. Hence, to perform forensic speaker verification, original voice is to be recovered from the disguised voice. Pitch disguise is one type of voice disguise which results in very low speaker recognition rate compared to other types of disguises. Original voice can be easily recovered from the pitch disguised voice if the level of disguise is known. Hence, this paper proposes a novel technique for finding the level of pitch disguise using Artificial Neural Networks (ANN) modelling. The performance of the system is measured in terms of accuracy and the proposed technique gives an accuracy of 96.23%