Formant dynamics of Spanish vocalic sequences in related speakers: A forensic-voice-comparison investigation

Abstract

This study investigates the dynamic acoustic properties of 19 vocalic sequences of Standard Peninsular Spanish, showing their potential for forensic voice comparison. Parametric curves (polynomials and discrete cosine transform) were fitted to the formant trajectories of the 19 Spanish vocalic sequences of 54 male speakers, comprising monozygotic (MZ) and dizygotic (DZ) twin pairs, non-twin brothers and unrelated speakers. Using the curve-fitting estimated coefficients as input to a multivariate-kernel-density formula, cross-validated likelihood ratios were calculated to express the probability of obtaining the observed difference between two speech samples under the hypothesis that the samples were produced by the same speaker and under the hypothesis that they were produced by a different speaker. The results show that the best-performing system is one that fuses the 19 vocalic sequences with a geometric-mean fusion method. When challenging the system with related speakers, the results show that MZ twin pairs affect performance but, more importantly, that non-twin sibling pairs can deteriorate performance too. This suggests that more investigations are necessary into a range of similar-sounding speakers beyond MZ twins. Several nurture aspects are highlighted as explanatory factors for the strikingly high similarity of a specific non-twin sibling pair.