الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 312 |  |  |
| | | from | 312 | | |
Abstract
Voice disguise is the most complex problem that faces the whole process of speaker identification; either in forensic case work or in biometric purposes, because it destroys the voice sample. Disguising a voice is particularly most relevant to forensic speaker identification FSI, which is concerned with identifying the unknown speaker in different forensic situations, wherein criminal intent occurs. In this situation, the process of speaker identification may be accomplished by humans alone, computers alone, or through a combination of both working together (Cain, Smrkovski & Wilson, 1990; Furui, 2008; and Hautamaki et al., 2017).
Voice disguise has two dimensional classifications; deliberate vs. non-deliberate, and electronic vs. non-electronic. Each type of the four resultant classifications has its own ways, tools and methods of evaluation. The present study is concerned mainly with electronic deliberate voice disguise, in which, an electronic tool (device, application, software, or etc.) is used to make deliberate modifications or intentional alterations of the voice of a perpetrator in order to conceal his/her identity. Most recent, there is tremendous development of internet, software and applications, which are available for free to change the voice of a speaker. For the criminals, such as these voice disguise tools can provide an effortless new interest for committing different crimes (Kunzel, 2000; Perrot et al., 2005; and Farrus, 2017).
Thus, voice disguise applications can affect lots of acoustical features of the original voice sample on segmental level and on supra-segmental level as well; which resulted in an unlimited number of new disguised voice samples. Therefore, the aim of the present experiment is to provide acoustical and perceptual insights into the effect of electronic deliberate voice disguise via a voice changer application. In addition to provide a detailed acoustic description of five different types of voice disguises, through 40 acoustic parameters. The present experiment includes 20 male speakers of Egyptian Colloquial Arabic, their normal voice samples are disguised electronically via voice changer android application, to generate five different types of disguises (child, giant, hoarse, old man & woman) of each normal voice sample. Moreover, there are 70 listeners who are volunteered for an online perceptual test that is designed to demonstrate the extent to which the listeners can verify the unknown speakers under the condition of different disguises. In addition to investigate their ability in identifying different types of voice disguises.
XXXV
The results indicate that, acoustically and statistically, there are 17 acoustic parameters have not significant differences between the normal & the disguised voice samples. These parameters are restricted in segment durations of the vowels and nasals, the third formant of the nasal /n/, and bandwidth of the first three formants of the vowels and nasals. They show a remarkable degree of stability and consistency under the different five types of disguises. However, there are 22 other acoustical parameters that have drastic changes according to the different five types of disguises. On the other hand, the perceptual results indicate that listeners can verify the unknown speakers under the condition of voice disguise by 83%; and can identify the type of the disguise by 53%. Moreover, listeners can correctly identify child, giant & hoarse disguise types by more than 85% of accuracy, while, they failed to identify old man and woman disguise types.