Forensic Automatic Speaker Recognition with dialectal speakers: A pilot study on the Taranto and Brindisi varieties

Francesco Sigona; Sonia D'Apolito; Cosimo Iaia; Barbara Gili Fivela; Mirko Grimaldi

doi:10.17469/O2110AISV000005

Autori

Francesco Sigona Dipartimento di studi umanistici, Centro di Ricerca Interdisciplinare sul Linguaggio, Università Del Salento, Lecce https://orcid.org/0000-0003-2939-0009
Sonia D'Apolito Dipartimento di studi umanistici, Università Del Salento, Lecce https://orcid.org/0009-0002-5593-5076
Cosimo Iaia Dipartimento di studi umanistici, Università Del Salento, Lecce https://orcid.org/0000-0002-4317-4540
Barbara Gili Fivela Dipartimento di studi umanistici, Università Del Salento, Lecce https://orcid.org/0000-0002-4694-8652
Mirko Grimaldi Dipartimento di studi umanistici, Università Del Salento, Lecce https://orcid.org/0000-0002-0940-3645

DOI:

https://doi.org/10.17469/O2110AISV000005

Parole chiave:

forensic voice comparison, likelihood ratio, reference population, spoken dialect, accuracy

Abstract

In the Forensic Speaker Recognition, the choice of the reference linguistic population plays a key role in evaluating the typicality of the voice recordings to be compared, and therefore the strength of the evidence, within the Likelihood Ratio framework. In the present study we carry out multiple comparison tests among voice recordings of the dialectal speech of Taranto and Brindisi varieties (Puglia, Italy), through an Automatic Speaker Recognition system, using two reference populations (databases): one of “standard” Italian, by professional speakers, and one of dialectal speakers of the same variety as the voices to be compared. The aim is to observe whether the accuracy of the recognition system improves if the dialect reference population is used instead of the Italian spoken one.