Speaker’s identification across recording modalities: A preliminary phonetic experiment

Sonia Cenceschi; Chiara Meluzzi; Nicholas Nese

doi:10.17469/O2107AISV000019

Authors

Sonia Cenceschi Digital Forensic Service, Department of Innovative Technologies, University of Applied Sciences and Arts of Southern Switzerland, Via Cantonale 2C, 6928 Manno, Switzerland https://orcid.org/0000-0002-4145-9593
Chiara Meluzzi Department of Humanities, Section of Linguistics, University of Pavia, corso strada Nuova 65, 27100, Pavia, Italy https://orcid.org/0000-0002-2291-006X
Nicholas Nese Department of Humanities, Section of Linguistics, University of Pavia, corso strada Nuova 65, 27100, Pavia, Italy

DOI:

https://doi.org/10.17469/O2107AISV000019

Keywords:

forensic phonetics, WhatsApp, forensic linguistics, environmental recordings, vowels’ formants, intra-speaker variation

Abstract

This work investigates how WhatsApp audio messages could be compared to high quality professional recordings and low quality ones in a forensic framework. A controlled experiment with 12 Italian students (6F, 6M) was performed in order to ascertain whether formants’ values of the three cardinal vowels /a/-/i/-/u/ will help in distinguishing the same speaker across three different recording modalities. Both unnormalized data in Hertz and normalized values (Lobanov and Bark) were compared across the male and female subsets. Results indicate that unnormalized data performed better than normalized ones, and that a qualitative investigation has to be combined with a quantitative one. This preliminary work opens the way to further investigations on the possibilities of WhatsApp audio messages for forensic purposes at the crossroads between linguistics and engineering.