Abstrakt
In everyday life, a speaker's voice characteristics change due to a range of factors such as speech style, affective states, daytime, or sickness; these shifts in the speaker's voice are possible due to extensive vocal tract plasticity and they often happen without the speaker's intention or even knowledge. This intra-speaker variability is, of course, of vital importance in the forensic phonetic context.
Another crucial aspect which can come into play in forensic voice comparison is intentional voice disguise, i.e., the speaker's deliberate attempt to conceal their identity by changing their voice characteristics. Some of the techniques used include placing a foreign object in front of or into their mouth (such as holding a tin can in front of their mouth as a resonator, covering their mouth with a cloth, or holding a pen between their front teeth; see Figueiredo & Britto, 1996), imitating a regional dialect or foreign accent, changing rhythmic characteristics of their speech, or trying to change their voice by articulatory or phonatory settings modifications.
Růžičková & Skarnitzl (2017) observed ways in which 100 Czech male speakers modified their voice when instructed to conceal their voice identity as much as they could in a manner of their own choice. The employed strategies differed among speakers, but in most cases, they were not very sophisticated, mostly including a change of a single parameter (predominantly speaking fundamental frequency, whose changes appeared in 70% of the speakers) or a combination of two.
This study presents our research focused on targeted voice disguise in five male speakers of Common
Czech. All of them are experienced voice users, trained in phonetics, and they were generally able to perform targeted voice manipulations. They were instructed to read a short text (the Czech translation of the Rainbow Passage) in their habitual voice, and several more times, each time performing a different resonance modification; the modifications were chosen mostly based on the SVPA scheme
(San Segundo & Mompean, 2017): (1) strong lip spreading, (2) lip rounding, (3) closed jaw, (4) open jaw, (5) palatalization, and (6) pharyngealization. Speakers could repeat any passage of the text multiple times in case they did not maintain the targeted voice manipulation. The recordings were performed in a sound-treated studio; they were later edited so as to contain the best realization of each sentence vis-a-vis the intended modifications.
The aim of this study is to describe in what exact ways the individual resonance modifications affect LTF and f0. This knowledge might be of help in cases where a speaker on a recording is apparently disguising their voice in a specific way and a prediction of the speech signal properties without the disguise is needed. This is a pilot study of our broader research on targeted voice disguise strategies' influence on instrumental and automatic speaker identification.