Examining long-term formant distributions as a discriminant in forensic speaker comparisons under a likelihood ratio framework
Erica Gold, Peter French, Philip Harrison · The Journal of the Acoustical Society of America · 2013
This study investigates the use of long-term formant distributions (LTFD) as a discriminant in forensic speaker comparisons. LTFD are the distributions calculated for all values of each formant for a speaker in a single recording. Spontaneous speech recordings from 100 male speakers of Southern Standard British English, aged 18–25 were analyzed from the DyViS Database (Nolan 2009). The recordings were auto-segmented to obtain a minimum of 50 s of vowels per speaker. The iCAbS (iterative cepstral analysis by synthesis) formant tracker was used to automatically extract and measure F1-F4 every 5 ms. To assess the evidential value of the LTFDs, likelihood ratios (LRs) were computed using a MatLab implementation of Aitken and Lucy’s (2004) Multivariate Kernel-Density formula (Morrison 2007). It was found that LTFD performs well overall, but much better with different speaker comparisons than same speaker comparisons (97.76 % compared to 78% of comparisons providing correct support; Cllr = 0.9072 and EER = 5.47%). LTFD appears to be a good discriminant to include in forensic speaker comparison analyses and offers the added attraction of avoiding potential correlation problems between vowel phonemes.