Forensic voice Comparison with Japanese Vowels Accoustics - a likelihood ration-based approach using segmental cepstra

Abstract

The suitability of vowel cepstral spectra for forensic voice comparison is explored within a likelihood ratio-based framework. Noncontemporaneous landline telephone recordings of 297 male Japanese speakers are compared using only two replicates each of their five vowels. 14 cepstrally-mean-subtracted LPC CCs from dc to 5 kHz are used as features. Multivariate likelihood ratios estimated for the 297 target- and 43956 nontarget trials give good results: an equal error rate of 0.28% and log likelihood ratio cost of 0.013. It is concluded that the approach has some merit.