Forensic Speaker Identification

A Likelihood Ratio-based Approach using Vowel Formants

Tony Alderman
Australian National University

This monograph describes an experiment in Forensic Speaker Identification, showing how speech samples from the same speaker can be discriminated from speech from different speakers with acoustic features commonly used in forensics. It also explains what is now considered the legally and logically correct approach to Forensic Speaker Identification, and presents data that can be used both in real casework and in further testing.

Forensic Speaker Identification is typically concerned with addressing the question of whether two or more speech samples have been produced by the same, or different, speakers. It is clear from recent research that the legally and logically correct way of doing this is by using a Bayesian Likelihood Ratio. The monograph explains what a Likelihood Ratio is; why its use is now considered correct; and how it can be used to successfully discriminate same-speaker pairs from different-speaker pairs.

The monograph shows how the Likelihood Ratio is a ratio of the probability of the evidence given a hypothesis (e.g. that the two samples are from the same speaker) to the probability of the evidence given a competing hypothesis (e.g. that the speech samples are from different speakers). This can be seen as a ratio expressing the similarity of the samples, divided by the typicality of the samples (i.e. how common these similarities are in the rest of the population). Since same-subject pairs are predicted by theory to have Likelihood Ratios greater than unity, and different-subject pairs are predicted to have Likelihood Ratios smaller, the Likelihood Ratio lends itself to use as a discriminant function to discriminate same-speaker from different-speaker speech samples. The extent to which this is possible is vital knowledge, given the legal evidentiary standards now accepted in the wake of the well-known Daubert rulings.

One stumbling block in the implementation of Bayesian Forensic Speaker Identification is the general lack of adequate background distributions for the assessment of the typicality of the similarities; that is, while two forensic speech samples may be similar, how common are the similarities in the general population?

Typically, one of the most important acoustic features used to compare forensic speech samples is vowel formants. These are the resonant frequencies of the speaker’s vocal tract when they are producing vowels. Bernard’s early study on the formants of male Australian English vowels, although now relatively old, provides potential background distribution data from a large number of speakers. The first goal of the monograph, therefore, is to describe, in adequate detail for forensic-phonetic investigation, the distributions of formant values for a subset of the vowels from the Bernard data set.

Many of the analytical methods used within Forensic Speaker Identification have an inherent assumption of normality for the distribution of the feature being analysed, whereas acoustic parameters from speech are often not normally distributed. Kernel Density estimation is one method which can take into account this non-normality. The second goal of the monograph, therefore, is to compare the performance of a ikelihood Ratio formula which assumes normality in the background distribution, with that of one using Kernel Density estimation.

The third aim of the research described is, through interpretation of the results of the discrimination tests, to add to the growing corpus of knowledge of the strength of various vowel-formant combinations as parameters for use in Bayesian Forensic Speaker Identification.

The tests are performed on data from eleven male speakers of Australian English, with non-contemporaneous samples, using formant values at target for their long monophthongal vowels to estimate the Likelihood Ratios. The results show that a formula assuming normality performs different-speaker comparisons more successfully than same-speaker, while a Kernel Density formula performs differently depending on the values chosen for within-speaker variance. Optimal results are found using the Kernel Density formula with within-speaker variance estimated from the test data. Suggestions for forensic practice are made with reference to these results. These include the possible use of both formulae for an analysis of different parameters, and further investigation of normality values and within-speaker variance on research outcomes.

ISBN 9783895867156. LINCOM Studies in Phonetics 01. 160pp. 2005.

The Use and Realisation of Accentual Focus in Central Catalan with a Comparison to English

Eva Estebas Vilaplana
Universidad Nacional de Educación A Distancia

This thesis investigates the use of accentuation to indicate broad and narrow focus in the Central dialect of Catalan. In response to a question such as “Who sang the song?” English speakers can say “JOE sang the song”, indicating the narrow focus of the answer by an accent on “Joe” and by the absence of accents on the following words. It has been claimed that in some Romance languages accent cannot be used to indicate focus in this way. Such languages choose instead to alter the grammatical structure of the response, for instance “The song, it was sung by Joe”. In this case, “Joe” is moved to an accentual prominent position.

Estebas-Vilaplana shows that in fact the “English-type” mechanism of achieving focus by re-structuring the prosody is fully available to a Central Catalan speaker and goes on to explore the details of its implementation. In doing this, she investigates quite a number of basic issues in the intonational analysis of Catalan within the Autosegmental-Metrical framework. She makes an interesting discovery that Central Catalan has a boundary tone aligned with the end of the word, and hence that a word level should be incorporated into the prosodic hierarchy. Eva Estebas-Vilaplana currently works at the Distance Learning University in Spain (UNED) as a lecturer in phonetics. She has published many papers on the intonation of Catalan and Spanish. She is the author of the book “Teach Yourself English Pronunciation”.

ISBN 9783895861376. LINCOM Studies in Phonetics 03. 247pp. 2009.

Ibn Sīnā

Risālah: ?ħasbāb ħuduwθ ?al-ħuruwf

A Treatise on Arabic Phonetics

Translation, Notes & Comments

Solomon I. Sara, S.J.
Georgetown University

Ibn Sīnā, also known in the West as Avicenna, is a 10th -11th century Persian Muslim philosopher and scientist. He was born around 980 C.E. / 370 H near Bukhara, in modern Uzbekistan, and died in 1037 C.E. / 428 H in Hamedan, in modern Iran. Among his philosophical and scientific works are kitāb ?al-šifā? ‘The book of healing’, his compendium of philosophy and science, and ?al-qānuwn fiy ?al-t`ibb ’The canon of medicine’ his compendium of medicine, which was translated into Latin and used in European universities as the primary medical source book for centuries.

The treatise under discussion is a unique linguistic-scientific treatise , written in Arabic, about Arabic phonetics from both scientific and linguistic perspectives. This thoughtful and innovative treatise went beyond the traditional impressionistic descriptions of the sounds of Arabic to include in its purview the three modalities of phonetics: the acoustic, the articulatory, and the anatomical/ physiological. In addition, it delved into comparative phonetics and natural parallels to the sounds of Arabic. It is of interest to linguists and phoneticians to note that this uncommonly broad and global perspective on phonetics would become common only in more recent research in and teaching of phonetics (19th century on). This treatise is a landmark in the scientific study and development of phonetics.

ISBN 9783929075915. LINCOM Studies in Phonetics 04. 200pp. 2009.

Experimental phonetics and sound change

Daniel Recasens, Fernando Sánchez Miret & Kenneth J. Wireback (eds.)
Universitat Autònoma de Barcelona; Universidad de Salamanca; Miami University

This book gathers some contributions from scholars working on the phonetic and phonological causes of sound change. Experimental evidence collected during the last decades calls for the need to build up better models of sound change which incorporate evidence from articulatory strategies, acoustic variation and perceptual categorization mechanisms. The papers collected in this book deal with the explanation of several sound changes, i.e., vowel shift and diphthongization, consonant voicing, assimilation, palatalization, vocalization and retroflexion, and with specific arrangements of places of articulation in sibilant inventories.

Contents:

Silvia Calamai & Irene Ricci, «Speech rate and articulatory patterns in Italian nasal-velar clusters».
Chiara Celata, «Rhotic retroflexion in Romance. Acoustic data for an articulation﷓driven sound change».
Juan Felipe García Santos, «Experimental analysis of some acoustically driven phonetic changes in Medieval Spanish».
Daniel Recasens & Aina Espinosa, «A perceptual analysis of the articulatory and acoustic factors triggering dark /l/ vocalization».
Joaquim Romero & Lucrecia Rallo, «An acoustic study of vowel shift in Majorcan Catalan». Fernando Sánchez Miret, «The effect of word final unstressed high vowels on stressed vowel duration and its consequences for metaphonic diphthongization in Southern Italian».
Marzena Żygis, «On changes in Slavic sibilant systems and their perceptual motivation». Kenneth Wireback, «A reexamination of the palatalization of Latin velar + dentoalveolar consonant sequences in the light of phonetic research».

ISBN 9783862880003 (Hardbound). LINCOM Studies in Phonetics 05. 140 pp. 2010.

Transcription of Intonation of the Spanish Language

Pilar Prieto & Paolo Roseano (eds.)
Institució Catalana de Recerca i Estudis Avançats and Universitat Pompeu Fabra; Universitat de Barcelona and Universitat Pompeu Fabra

The main purpose of this book is to take a step forward in the development of a consensus-based transcription system for Spanish intonation within the Tones and Break Indices (ToBI) framework by taking into account data from different dialectal varieties. The book comprises ten chapters which describe the basic intonational patterns of the following dialects of Spanish: Castilian, Cantabrian, Canarian, Dominican, Puerto Rican, Venezuelan Andean, Ecuadorian Andean, Chilean, Argentinian and Mexican Spanish.

Each chapter offers a Sp_ToBI analysis of the basic intonational contrasts found in these dialects that is based on a wide range of data including a variety of statements, yes-no questions, wh- questions, echo questions, as well as vocatives and imperatives. One of the advantages of the present work is that all chapters share a common methodology in data collection, thus permitting a precise comparison of the resulting intonation contours and their analyses. Digital recordings of the actual data on which the chapters are based are available at the website of the Interactive Online Atlas of Spanish Intonation (Atlas interactivo de entonación del español, accessible at http://prosodia.upf.edu/atlasentonacion/). We hope that the work presented here will serve as an empirical groundwork for the establishment of a standard, comprehensive system for transcribing the intonation of the Spanish language.

Contents:

Introduction
Pilar Prieto

Castilian Spanish Intonation
Eva Estebas-Vilaplana and Pilar Prieto

Cantabrian Spanish Intonation
M.ª Jesús López-Bobo and Miguel Cuevas-Alonso

Canarian Spanish Intonation
Mercedes Cabrera Abreu and Francisco Vizcaíno Ortegarie

Dominican Spanish Intonation
Erik W. Willis

Puerto Rican Spanish Intonation
Meghan E. Armstrong

Venezuelan Andean Spanish Intonation
Lluïsa Astruc, Elsa Mora and Simon Rew

Ecuadorian Andean Spanish Intonation
Erin O'Rourke

Chilean Spanish Intonation
Héctor Ortiz, Marcela Fuentes and Lluïsa Astruc

Argentinian Spanish Intonation
Christoph Gabriel, Ingo Feldhausen, Andrea Pešková, Laura

Colantoni, Su-Ar Lee, Valeria Arana and Leopoldo Labastía

Mexican Spanish Intonation
Carme de-la-Mota, Pedro Martín Butragueño and Pilar Prieto

ISBN 9783862901845 (Hardbound). LINCOM Studies in Phonetics 06. 380pp. 2010.

Perception of Castilian Spanish Intonation

Implications for Intonational Phonology

Timothy L. Face
University of Minnesota

Perception of Castilian Spanish Intonation: Implications for Intonational Phonology presents four perception experiments on the intonation of Castilian Spanish, two dealing with the declarative vs. absolute interrogative distinction and two dealing with the broad focus vs. narrow focus distinction in declaratives. These experiments provide insight into the intonational system of Castilian Spanish that goes beyond what is possible through studies of speech production. For both distinctions investigated, the aim of the experiments is to determine which of the multiple intonational cues lead listeners to perceive sentence type or focus type. The experimental results not only lead to a better understanding of how intonation communicates sentence type and focus type in Castilian Spanish, but also have implications for the phonological analysis of Castilian Spanish intonation as well as for the Autosegmental-Metrical theory of intonational phonology. With respect to the analysis of Castilian Spanish intonation, the experimental results present a challenge to the common analysis that F0 rises that begin near the onset of the stressed syllable, but which differ based on the alignment of the F0 peak, result from two phonologically distinct pitch accents.

With respect to the Autosegmental-Metrical theory of intonational phonology, the results present a challenge to its limited ability to account for linguistically meaningful distinctions in pitch scaling beyond the High vs. Low tone distinction and also for the often assumed compositional approach to intonational meaning. For each of these challenges, a new proposal is offered to more adequately account for the data.

Timothy L. Face is Associate Professor of Hispanic Linguistics at the University of Minnesota. His research focuses on phonetics and experimental phonology, and especially on Spanish prosody. He has published extensively on Castilian Spanish intonation, including two previous books, Intonational Marking of Contrastive Focus in Madrid Spanish (2002) and The Intonation of Castilian Spanish Declaratives and Absolute Interrogatives (2008). In addition, he is author of Guide to the Phonetic Symbols of Spanish (2008), editor of Laboratory Approaches to Spanish Phonology (2004) and founding editor of the journal Studies in Hispanic and Lusophone Linguistics.

ISBN 9783862880461. LINCOM Studies in Phonetics 07. 114pp. 2011.

Quantitative approaches to problems in linguistics

Studies in honour of Phil Rose

Cathryn Donohue, Shunichi Ishihara, William Steed (editors)

This collection of eighteen papers is a festschrift volume to honour Phil Rose on the occasion of his recent retirement from the Australian National University. Phil’s work in both tonal acoustics and forensic voice comparison has always improved upon the accepted standards through the insistence on rigorous statistical analyses and the introduction of techniques not previously employed in the field.

The contributors to the volume take the type of quantitative approach espoused by Phil and apply it to problems in linguistics, presenting statistically-founded solutions. The first six papers cover tones: the environments in which they occur, tonal acoustics of North-central Vietnamese, and of two Japanese varieties, the application of tonal acoustics to Qingtian morphotonemics, tonal identification in Fuzhou, and tone alternations in Ugong. Additionally there are three papers on vowel changes in Chinese, Japanese phonological acquisition, and the phonological systems of Australian languages.

The second section groups together six papers on forensic voice comparison (FVC), some using Phil’s general Bayesian approach to FVC. These include a tutorial on linear-scaling effects of phonetic context on vowel formants, the implications for FVC of formant frequencies over the phone in Japanese, an approach to automatic speaker identification using the magnitude and phase spectra of inverse-filtered voice speech, FVC in Chinese using /iau/, and a final paper applying the general approach to the challenges of Language Analysis used to Determine the Origin of people applying for refugee status (LADO).

The final section consists of three papers on topics outside these areas but which have nonetheless been influenced by Phil and his work. They include a paper on UG and variation in expression, another on contextualizing the Old Javanese influence on Old Japanese, as well as a lexical-conceptual analysis of ‘eating’ and ‘drinking’ in two varieties of Chinese.

List of Contributors

Ann Kumar, Cathryn Donohue, Shunichi Ishihara: Phil Rose: A short biography

Cathryn Donohue, Shunichi Ishihara, William Steed: Many voices, many tones

Part I. Tones and Acoustic Phonetics

Mark Donohue: The shape and spread of Tone

William Steed: Qingtian Wu lexical tone sandhi: Voiceless depressors and allotones

Koichi Honda: The realisation of the stopped tone in North-Central Vietnamese

David Bradley: Tone alternations in Ugong (Thailand)

Cathryn Donohue: The role of contour and phonation in Fuzhou tonal identification

Shunichi Ishihara: Osaka and Kagoshima Japanese citation tone acoustics: A linguistic tonetic comparative study

Xiaonong Zhu: Off-the-Chart vowel changes in Chinese

Takako Toda: The critical period hypothesis and phonological acquisition of Japanese

Andrew Butcher: On the phonetics of long, thin phonologies

Part II. Forensic voice comparison

Frantz Clermont: Linear-scaling effects of phonetic context on vowel formants: A tutorial

Michael Carne: Japanese formant frequencies in mobile phone transmission: Implications for Forensic Voice Comparison

Javier Franco-Pedroso, Joaquin Gonzalez-Rodriguez, Javier Gonzalz-Dominguez, Daniel Ramos: Fine-grained automatic speaker recognition using cepstral-trajectories in phone units

Michael Wagner: Automatic speaker identification using the magnitude and phase spectra of inverse-filtered voiced speech

Cuiling Zhang, Geoffrey Morrison, Tharmarajah Thiruvaran: Forensic voice comparison using Chinese /iau/

Helen Fraser: Bayes and beyond: The complex challenges of LADO and their relevance to forensic speaker comparison

Part III. Bayes and beyond

Avery D Andrews 3rd: UG and variation in expression

Ann Kumar: ‘Humble auxiliaries’ in Old Japanese: Javanese derivations, context, and significance

Zhengdao Ye: Eating and drinking in Mandarin and Shanghainese: A Lexical-Conceptual analysis

Index

ISBN 9783862883844 (Hardbound). LINCOM Studies in Phonetics 08. 300pp. 2012.

Phonetische und linguistische Prinzipien des forensischen Stimmenvergleichs

Michael Jessen
Kriminaltechnisches Institut des Bundeskriminalamtes Wiesbaden

Bei einem forensischen Stimmenvergleich werden für die Zwecke der Beweisführung vor Gericht Audioaufzeichnungen der Stimme eines Täters mit denen der Stimme eines Verdächtigen verglichen. Auf Grundlage einer solchen stimmenvergleichenden Analyse wird auf die Frage eingegangen, mit welcher Wahrscheinlichkeit Täter und Verdächtiger identisch oder nicht-identisch sind. Phonetik und Linguistik sind die wesentliche wissenschaftliche Basis, auf der ein forensischer Stimmenvergleich fußt.

In diesem Buch werden die Prinzipien des forensischen Stimmenvergleichs erklärt und auf die spezifischen Schwierigkeiten hingewiesen, die forensisches Audiomaterial mit sich bringen kann (eingeschränkter Frequenzgang, reduzierte Dauer, lautes Sprechen usw.). Im Detail wird gezeigt, in welcher Weise die Methoden und Erkenntnisse der Phonetik und Linguistik in dem Prozess der forensischen Analyse zum Tragen kommen. Hierbei werden praktische Beispiele gegeben und es wird besonders auf neuere Forschungsergebnisse eingegangen, die im Labor des Autors und in der nationalen und internationalen Sprechererkennung erzielt wurden. Hintergrundstatistiken über Merkmalsgruppen wie Grundfrequenz, Formantenfrequenzen und Sprechtempo, die dabei zur Sprache kommen, sind nicht nur für die Forensik relevant, sondern auch für andere Bereiche – wie zum Beispiel die klinische Linguistik – in denen es auf Normdaten ankommt, die anhand einer großen Anzahl von Sprechern erhoben wurden.

Michael Jessen ist Experte für Sprechererkennung und Tonträgeranalyse am Kriminaltechnischen Institut des Bundeskriminalamtes in Wiesbaden. Er studierte Linguistik mit Schwerpunkt Phonetik und Phonologie an der Universität Bielefeld und der Cornell University und war bis 2001 wissenschaftlicher Mitarbeiter am Institut für Maschinelle Sprachverarbeitung der Universität Stuttgart. Seit 2009 ist er Co-Editor der Fachzeitschrift The International Journal of Speech, Language and the Law.

ISBN 9783862883219. LINCOM Studies in Phonetics 09. 247 S. 2012.

Dutch & Afrikaans Pronunciation & Accents

Luciano Canepari & Marco Cerini
University of Venice

This book applies the principles of Natural Phonetics & Tonetics to describe the pronunciation of both Dutch and Afrikaans, including intonation, in a precise way never found in earlier treatises. It includes an introduction to the Natural Phonotonetics Method, which can be used for other languages, as well. In addition to Afrikaans pronunciation (with some variants), five kinds of Dutch pronunciation are systematically described: International, Neutral & Mediatic Flemish, and Neutral & Mediatic Netherlandic. Also 22 European regional accents are described, more concisely but still accurately and always with intonation, too. The pronunciation of Dutch by people from South Africa, the Netherlands Antilles and Surinam is given, as well. The synopses of Old and Middle Dutch are provided, too. The book is completed by a mini-phono-dictionary containing the transcription of almost 2000 place, family and person names.

Luciano Canepari (Natural Phonetics & Tonetics, University of Venice, Italy, originally trained in the British phonetic tradition, later developed the Natural Phonotonetics Method) and Marco Cerini (PhD in Natural Phonetics & Tonetics, applied to Mandarin Chinese pronunciation, University of Venice, Italy).

Contents:

1. Foreword
2. Pronunciation & Phonetics
3. The phono-articulatory apparatus
4. The classification of sounds
5. Vowels & vocoids
6. The vowels & diphthongs of international Dutch
7. Consonants & contoids
8. The consonants of international Dutch
9. Structures
10. Intonation
11. Some texts in phonotonetic transcription
12. Neutral Flemish-Dutch pronunciation
13. Mediatic Flemish-Dutch pronunciation
14. Neutral Netherlandic-Dutch pronunciation
15. Mediatic Netherlandic-Dutch pronunciation
16. Regional Dutch accents (with five maps)
17. Non-European Dutch accents
18. Afrikaans pronunciation
19. Mini-phono-dictionary
20. Old-Dutch & Middle-Dutch synopses
21. Annotated bibliography

ISBN 9783862884957 (Hardbound). LINCOM Studies in Phonetics 10. 212pp. 2013.