Original scientific paper
Forensic Phonetic identification and linguistic analysis of the speaker
Gordana Varošanec Škarić
orcid.org/0000-0002-0864-4252
; Faculty of Humanities and Social Sciences
Gabrijela Kišiček
orcid.org/0000-0002-5055-9609
; Faculty of Humanities and Social Sciences
Abstract
Upon the request of the County Court of Zagreb, on the basis of the recordings obtained
by the police eavesdropping of the GSM device, the identification of a male voice was to be
performed. The male voice from the recording No. 1 – duration of 2.16 min (CD 1) had to be
compared with the male voice from the recording No. 2 – duration 7.39 min. The whole material
was phonetically transcribed for the purpose of examining regional pronunciation, dialect and
idiolect. For the purpose of sound analysis, dubbing was performed as well. Linguistic analysis
has shown the difference in the use of idiosyncratic words, as well as the difference on the
phonological level. For instance, recognized speaker used word jaran 69 times (which is 170
times more that in the usage of Croatian language). In the sample of two recordings with the
recognized speaker, idiosyncratic meaningful words such as [jǎran; bolan] show significantly
more frequent absolute and relative appearance when compared with the appearance in Croatian
language. Those words didn’t appear in the sample of the second speaker, but the recordings
had some other words that coincided such as [razǔmiš; pajdo]. Although the phonetic and
linguistic analyses differ, they are tightly connected. According to the forensic phonetics protocol
(SPID, Hollien 2002, Varo{anec–[kari}, 2008), auditory perception analysis of speech show
high possibility of male person identification, i.e. experts’ recognition was 96% in range from94% to 98%. Recognized speaker belongs to the group of stakavian ikavian Bosnian speeches:
syntactically and morphologically that is eastern Bosnian variant. Phonetically, on the suprasegmental level of accent realization, the
recognized speaker uses marked accents of the eastern stakavian variant (longer long–rising
accent and after–stress length), he is shifting falling accents from the meaningful words to the
proclitic, and on the segmental level his pronunciation of vowels is typical for stokavian dialect.
Vowels are realized more forward and closed than those of cardinal vowels of Croatian Received
Pronunciation, reduction and after–stress omitting of vowels appear. Phonetic description of
voice shows slightly higher voice, narrow jaw aperture, dentalization, and modal phonation type.
Furthermore, formant analysis of all vowels was made for both speakers (F2, F3), while vowel
/a/ provided the analysis of F1 as well, because its average values are around 800 Hz and is not
affected by the distortion of transmission. Formant analysis based on the stresses vowels has
shown the difference (z–test) when compared with standard Croatian Pronunciation (for lower
F1 /a/: p = 0,001, F2 higher for /a/, /e/, /o/, /u/) and in comparison with the other speaker F2
was higher for vowel /i/ (p = 0,08), F3 lower for vowel /o/ (p < 0,001). Long average spectra
of speech that give information about voice timbre are different for those two speakers. Typical
distortions in lower and higher parts of the spectrum during the transmission via GSM device
were taken into consideration. According to Harmegnies (1995) similarity index (R) was calculated for two spectra from
different recordings of the same identified male speaker was 0.91 and dissimilarity index (SDDD)
was 0.96. In pair with unknown speaker R was in range from 0.75 to 0.88 and SDDD in range
from 3.32 to 4.50. According to total results it can be concluded that male speaker was identified
with a high rate of identification probability.
Keywords
speaker identification; formant frequencies; speaker identification protocol; forensic phonetics; forensic linguistics
Hrčak ID:
83040
URI
Publication date:
11.6.2012.
Visits: 3.352 *