Likelihood-ratio forensic voice comparison using parametric representations of the formant trajectories of diphthongs

Abstract

Non-contemporaneous speech samples from 27 male speakers of Australian English were compared in a forensic likelihood-ratio framework. Parametric curves (polynomials and discrete cosine transforms) were fitted to the formant trajectories of the diphthongs /a/I, /eI/, /o[see text]/, /a[see text]/, and open /[see text]I/. The estimated coefficient values from the parametric curves were used as input to a generative multivariate-kernel-density formula for calculating likelihood ratios expressing the probability of obtaining the observed difference between two speech samples under the hypothesis that the samples were produced by the same speaker versus under the hypothesis that they were produced by different speakers. Cross-validated likelihood-ratio results from systems based on different parametric curves were calibrated and evaluated using the log-likelihood-ratio cost function (C(llr)). The cross-validated likelihood ratios from the best-performing system for each vowel phoneme were fused using logistic regression. The resulting fused system had a very low error rate, thus meeting one of the requirements for admissibility in court.

Cross-dialectal variation in formant dynamics of American English vowels

“…Having established the types of variation in formant trajectories that can be expected in cross-dialectal data in terms of TL, directionality, and curvature, a refinement of the current measures will be undertaken in order to address the changes in the direction of formant movement. In particular, parametrization procedures can be used ͑e.g., Harrington, 2006;Harrington et al, 2008;Hillenbrand et al, 2001;Morrison, 2009;Zahorian and Jagharghi, 1993͒ in order to model the various trajectory shapes.…”

Formant trajectories in linguistic units for text-independent speaker recognition

“…This variable-length segment is duration equalized to a number of frames equivalent to 250 ms, following results in previous studies [3] [10]. Finally, those three temporal trajectories within a unit are coded by means of a fifth order DCT and its coefficients concatenated, yielding our final 3x5=15 fixed-dimension feature vector for each linguistic unit.…”

Semi-automatic speaker verification system

“…Обобщенное решение о сходстве либо различии голосов дикторов принимается автоматически, в резуль-тате фузирования используемых методов. Более ранние исследования в области криминалистического распознавания дикторов по голосу были сосредоточены на статистическом анализе распределения таких акустических и просодических признаков, как частота основного тона [3, 5, 6], частоты формант [7][8][9] и темпоральные супрасегментные характеристики [10,11]. До недавнего времени относительно мало вни-мания уделялось исследованию специфичных для речи диктора статистик длительностей фонем.…”