Intra- and inter-speaker variability in duration and spectral properties of English /s/

Abstract

This study investigates the speaker-specificity of acoustic characteristics of the English fricative /s/ and contributes background population statistics for use in forensic speaker comparison work. The intra- and inter-speaker variability in duration and spectral properties of /s/ was investigated in data from 30 young adult male speakers of Cambridge and Leeds English. Read speech was used in the present study to allow direct comparison across speakers. Segment duration was normalized for speaking rate. Spectra were filtered at 4 kHz in order to explore speaker discrimination performance at settings mimicking the bandpass filter effect of telephone transmission. Additional filters were applied at 8, 16, and 22.05 kHz to investigate discrimination with data from various frequency ranges. Spectral measures were calculated from a 40-ms window centered on the midpoint of each token. Although mean values display relatively little inter-speaker variation, the individuals at the extreme high and low ends of the distributions may be the best discriminated, particularly those at the extremes on more than one parameter. Discriminant analyses were conducted to determine the most speaker-specific predictors; relative performance was compared across the four filter conditions. The discriminatory ability of these parameters will also be presented using a likelihood ratio framework.