Testing realistic forensic speaker identification in Japanese: a likelihood ratio-based approach using formants

Abstract

IllI would also like to thank Drs. 'Spike' Barlow and Frantz Clermont of the Australian Defence Force Academy for giving me many useful comments in seminars. I must thank my proofreaders, Sue Novak and Andrew Lim who went through bits of my earlier draft, Pascale Jacq who helped me with the final check of my English usage, references, tables and figures etc, and Ingrid Singh who has been reading my extremely messy drafts over four years from the earliest stage of this research. Ingrid was so generous as to lie to me that she enjoyed reading the drafts and was eagerly waiting for more. What a wonderful friend.Last but not least, I would like to thank my family and friends who provided moral support, which was invaluable at various stages during the construction of this PhD thesis. I would like to thank my mum, dad and sister, Shoko, for being there for me at the toughest time (so far!) in my life and backing me up in their own individual very different ways. I am also very grateful to my grandmother in Japan and grandfather somewhere up there for their love and understanding. The fact that they feel pride for me, instead of disappointment for staying at university for all these years has given me a great deal of strength.All my friends, both in Japan and Australia, have been great. Everyone did a wonderful job of making me laugh, cheering me up, and calming me down when I panicked, but I would especially like to thank Andrew and Ingrid. Andrew made me smile (and laugh) at least ten times a day, even when I was exhausted and miserable, which is impressive work. Ingrid has been there for me throughout entire my PhD period. I believe that my life as a PhD candidate would have been so much tougher without her friendship.

Proposta de Construção de um Banco de Dados de Amostras de Fala para Uso Forense em um Arcabouço Bayesiano

“…Parâmetros não vocálicos também têm sido estudados segundo o arcabouço bayesiano. Assim, frequência fundamental, VOT (Voice Onset Time), nasais, laterais, fricativas, taxa de articulação, espectro de longo termo, clicks, entre outros parâmetros, têm recebido atenção de vários estudiosos [40,[49][50][51][52].…”

Issues and opportunities: The application of the numerical likelihood ratio framework to forensic speaker comparison

“…Much of the early LR literature used Lindley to compute strength of evidence in order to assess the discriminatory potential [21,27,65]. The appropriateness of Lindley for modelling traditional acoustic-phonetic data (primarily vowel formants) has also been assessed directly in [66,67].…”

“…Non-contemporaneity is defined here as speech samples that have been recorded in different sessions separated by a non-specific amount of time. This default H d has been reflected in the majority of LR-based studies using traditional acoustic-phonetic parameters [32,21,42,66].…”

Forensic voice comparison and the paradigm shift

“…The first such implementation was in Yuko Kinoshita's PhD dissertation completed in 2001 [62]. In 2002 and 2003 Philip Rose published a book and a book chapter on likelihood-ratio forensic voice comparison, the first aimed primarily at phoneticians [16], and the second aimed primarily at the legal community [32].…”