Detecting voice disguise from speech variability : analysis of three glottal and vocal tract measures

Abstract

The deliberate attempt by speakers to conceal their identity (voice disguise) presents a challenge for forensics and for automated speaker identification systems. Using a database of natural and disguised voices of three professional voice artists, we build on earlier findings in [Amin et al., 2014] by exploring how certain glottal and vocal tract measures, such as glottal timing (Open Quotient) and vowel formants are manipulated by the artists to create novel voice identities. We also investigate whether there are any features from these measures that can be useful for discriminating natural and disguised voices. As expected, variation in Open Quotient was speaker-dependent, and corresponded closely to social attributes (i.e., age) of the voice identities involved. By modelling the overall variability of speakers in the vowel space, we propose a new method for automatically classifying natural and disguised voices. The proposed method is found to outperform several state-of-the-art methods.