Estatísticas populacionais da frequência fundamental do português brasileiro para uso em fonética forense

Abstract

This research presents the fundamental frequency (f0) statistics of 100 male speakers of Brazilian Portuguese (BP) who produced samples of semi-spontaneous speech and sentence reading. The analyzed speech samples are part of the Forensic Corpus of Brazilian Portuguese (CFPB), provided by the Audiovisual and Electronics Sector (SEPAEL), of the National Institute of Criminalistics (INC) of the Brazilian Federal Police. The demographic characteristics of the 100 samples follow, as far as possible, the distribution of the Brazilian population in terms of age group and distribution among the five major Brazilian regions according to the 2020 National Household Sample Survey (PNAD). Excerpts with duration of liquid speech between 30 and 45 seconds were analyzed. A set of eight statistical measures were reported for each f0 sample: mean, median, mode, baseline value, standard deviation, coefficient of variation, asymmetry and kurtosis, which were presented in the form of graphs, tables, histograms and scatter plots. It was observed that both the elocution style and the conversion of f0 values ​​from the Hertz scale to the semitone scale increase the proportion of distributions with values ​​that adhere to the normal distribution. Regarding the regional variation, the results suggest the absence of an important influence of the geographic region on the value of the collected statistical estimators. The results indicate convergence with the literature and are discussed in relation to the phonetic-forensic implications, and can be used by experts to calculate likelihood ratio values ​​in real scenarios in forensic voice comparison tasks.