Likelihood Ratio-based Forensic Voice Comparison on L2 speakers: A Case of Hong Kong native male production of English vowels
Year: 2015
Type: conference-paper
Abstract: The performances of two different procedures for calculating \nthe likelihood ratio (LR) for forensic text comparison (FTC) are \nempirically compared. One is the multivariate kernel density (MVKD) \nprocedure with so-called lexical features. The MVKD procedure has \nbeen successfully applied to various types of evidence, including texts. \nThe other is the procedure based on character N-grams. N-gram is a \nwidely-used, robust probabilistic language model. The effectiveness of \ncharacter N-grams has been reported in authorship analysis, however, to \nthe best of my knowledge, it has not yet been applied to LR-based FTC. \nIn this study, the log-likelihood-ratio-cost (Cllr), which is an appropriate \nassessment metric for LR-based systems, is used to assess the \nperformance of the two procedures. It will be reported that the MVKD \nprocedure outperforms the character N-gram procedure. Through the \ncomparison of the two procedures, this study also demonstrates how the \nweight of evidence (= LR) can be estimated from text messages.
Authors: D. Frost, Shunichi Ishihara
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W2563484817