Forensic Voice Comparison

Selected Publications and Other Documents

Rose, P., & Winter, E. (2010) Traditional forensic voice comparison with female formants- Gaussian mixture model and multivariate likelihood ratio analyses. Manuscript submitted for publication.

The first likelihood ratio-based forensic voice comparison on female voices, and the first forensic use of Gaussian mixture models on traditional features, are described. A GMM-UBM LR-based comparison is performed on the first three formants of the five long monophthongs of 20 General Australian English female speakers in non-contemporaneous recordings separated by one to five weeks. Comparison with logistic-regressively fused multivariate likelihood ratios from the same data shows both systems perform well, but the latter is superior in both EER and log likelihood ratio cost.

Morrison, G. S. , et al. (2009) Making demonstrably reliable forensic voice comparison a practical everyday reality in Australia. Linkage Project Application.

To make forensic voice comparison a practical everyday reality in Australia, for use in police investigations and for presentation of evidence in court, forensic scientists must have a practical and demonstrably reliable forensic-voice-comparison system including a large representative database of Australian English voices. In collaboration with Australian police forensic laboratories and other partners we will develop and test such a system, improving on existing systems by combining the strengths of acoustic-phonetic and automatic approaches. The system will comply with the strictest international standards for the admissibility of scientific evidence in court, using the same evaluative framework as is applied to DNA.

Rose, P. (2009) Evaluation of disputed utterance evidence in the matter of David Bain’s retrial. Report prepared for court case, 23 February2009

An acoustic and auditory-phonetic analysis of the data in Bain’s emergency call shows strong evidence in support of the hypothesis that the disputed utterance begins with Bain saying I can’t, and against the hypothesis that it begins with I shot. It is not convincing to argue that the disputed utterance was not speech.

Interview on 3news.co.nz Campbell Live

Morrison, G. S. (2009) The place of forensic voice comparison in the ongoing paradigm shift. The 2nd International Conference on Evidence Law and Forensic Science Conference Thesis (Vol. 1, pp. 20–34). Beijing, China: The Key Laboratory of Evidence Science of the Ministry of Education (The Institute of Evidence Law and Forensic Science, China University of Political Science and Law).

We are in the midst of what Saks & Koehler (2005) have called a paradigm shift in the evaluation of evidence in the forensic comparison sciences. This is a shift to requiring that the evaluation of forensic evidence actually be scientific, including that the reliability of methodologies be testable, and requiring that forensic evidence be evaluated and presented to the courts in a logically correct manner. Reliability was a primary concern in the US Supreme Court’s Daubert decision in 1993. The logically correct evaluation and presentation of the weight of forensic evidence was a primary concern in a number of court cases in the United Kingdom including the Appeal Court of England and Wales’ 1996 ruling on the presentation of DNA evidence in R v Doheny & Adams. In the US National Research Council’s report to Congress released in February 2009 current practice in the evaluation of nuclear DNA evidence is held up as a model to emulate, and current practices in other branches of forensic science are subject to sometimes severe criticism. In the present paper I examine the place of forensic voice comparison in the ongoing paradigm shift. Over the last decade a small number of forensic-voice-comparison researchers have been working in the post-shift paradigm. They have adopted the likelihood-ratio framework for the evaluation of forensic evidence, the same framework as used in DNA analysis. I provide a brief description of the likelihood-ratio framework, followed by a brief history of the adoption of the likelihood-ratio framework for forensic voice comparison by the research community, and by the forensic practitioner, law-enforcement, and judicial communities.

A revised version of this paper is: Morrison, G. S. (2009) Forensic voice comparison and the paradigm shift. Science & Justice, 49, 298–308.

Morrison, G. S. (2009) Comments on Coulthard & Johnson’s (2007) portrayal of the likelihood-ratio framework.

In their recent introduction to forensic linguistics, Coulthard & Johnson (2007) include a portrayal of the likelihood-ratio framework for the evaluation of forensic comparison evidence (pp. 203–207). This portrayal includes a number of inaccuracies. The present letter attempts to correct these.

A revised version of this paper is published as Morrison, G. S. (2009). Comments on Coulthard & Johnson’s (2007) portrayal of the likelihood-ratio framework. Australian Journal of Forensic Sciences, 41, 155–161.

Zhang, C. , & Morrison, G. S. (2008) Forensic comparison of Chinese female voices. Successful application for International Association for Forensic Phonetics and Acoustics Research Grant. Research funded: 2009–2010.
- Research on forensic voice comparison has primarily focussed on male voices, which is understandable given that most casework deals with male voices; however, in the Principal Applicant’s casework experience approximately 20% of cases involve female voices. There is therefore a need to conduct research on the effectiveness of forensic techniques when applied to female voices. The present study examines the effectiveness of forensic voice comparison based on parametric curves fitted to the formant trajectories of some frequently-occurring Chinese diphthongs and triphthongs produced by female speakers. The aims of the research are:

Rose, P. (2009) Response to the Law Commission of England and Wales Consultation Paper 190: The admissibility of expert evidence in criminal proceedings in England and Wales: A new approach to the determination of evidentiary reliability.

Morrison, G. S. (2009) Response to the Law Commission of England and Wales Consultation Paper 190: The admissibility of expert evidence in criminal proceedings in England and Wales: A new approach to the determination of evidentiary reliability.

Zhang, C. , & Morrison, G. S. (2008) Forensic comparison of Chinese female voices. Successful application for International Association for Forensic Phonetics and Acoustics Research Grant. Research funded: 2009–2010.

Rose, P., & Morrison, G. S. (2008) A response to the UK postion statement on forensic speaker comparison. Manuscript.

We have identified three basic weaknesses in the proposal. Firstly, the apparent treatment of speech as if its features were discrete and invariant, like DNA, has the result that the approach would be very difficult to implement. Secondly, we pointed out the inconsistencies in first prohibiting but then allowing probability of hypothesis, given evidence statements; and proposing distinctiveness statements while acknowledging the absence of information upon which to base them. Thirdly, the absence of a way of relating the consistency and distinctiveness assessments does not help the trier-of-fact to interpret them. We would reject the claim in the Statement’s foreword that the UK Framework is “at a conceptual level, identical to that used nowadays in the presentation of DNA evidence” (p. 138). Consequently, we would also argue that the UK Position Statement has not achieved its goal, however laudable, of “bring[ing] the field [of forensic voice comparison] into line with modern thinking in other areas of forensic science” (p. 137). It is the use of likelihood ratios that characterises modern thinking on the evaluation of forensic comparison evidence.We would encourage forensic-voice-comparison researchers and practitioners world-wide to rapidly move towards adopting quantitative likelihood-ratio statements as standard.

A revised version of this paper is: Rose, P. & Morrison, G. S. (2009) A response to the UK position statement on forensic speaker comparison. International Journal of Speech, Language and the Law, 16, 139–163.

Rose, P., & Morrison, G. S. (2008) Una respuesta a la declaración de posicionamiento británico sobre la comparación forense de locutores. Translation by J. J. Lucena Molina. Unpublished Manuscript.

Hemos identificado tres básicos puntos débiles en la propuesta. Primeramente, la aparente consideración del habla como poseedora de propiedades discretas e invariantes, a semejanza del ADN, tiene como resultado que el enfoque sería muy difícil de implementar. En segundo lugar, subrayamos la incoherencia de, por un lado, prohibir las probabilidades de las hipótesis, dada la evidencia y, por otro, permitirlo; así como proponer afirmaciones sobre peculiaridad mientras se reconoce la ausencia de información sobre las que basarse. En tercer lugar, la ausencia de una forma de relacionar las afirmaciones sobre consistencia y peculiaridad no ayuda al Jurado o al Tribunal a interpretarlas. Rechazaríamos lo que se defiende en el prefacio de la Declaración de que el marco de la Declaración es “en el nivel conceptual, idéntico al que se utiliza hoy día en la presentación de la evidencia de ADN” (página 138). Consecuentemente, argumentaríamos que la Declaración de posicionamiento del Reino Unido no ha alcanzado su objetivo, sin embargo laudable, de “… llevar la disciplina [de comparación forense de voz] dentro del pensamiento moderno en otras áreas de la ciencia forense” (página 137). Es la relación de verosimilitud la que caracteriza el moderno pensamiento sobre la evaluación de la evidencia de comparación forense. Nosotros animaríamos a los investigadores y expertos en comparación de voz forense a que rápidamente se adaptaran a realizar afirmaciones de relación de verosimilitud cuantitativas como estándar.

Rose, P., & Morrison, G. S. (2008) 对英国关于法庭话者比较的立场声明的回应. Translation by Cuiling Zhang 张翠玲. Unpublished Manuscript

我们明确了英国提出的建议中的三个根本缺陷。首先，将语音特征作为离散的和不变的特征来对待，就象DNA一样，其结果是很难将该方法付诸实现。其次，我们指出了文章前后的矛盾之处，即先是禁止后又允许使用证据的假设概率，承认缺少支持信息却又使用特殊性表述。第三，缺少关于一致性和特殊性的评价方法并没有帮助事实裁定者对其进行解释。我们反对英国体系在声明的前言中宣称的：“在概念水平上，与当前提供DNA证据时使用的方法一致”(p. 138) 的观点。最后，我们还要指出的是，英国立场声明并没有达到它的目标。但是，值得称赞的是，“它通过现代思想将法庭语音比较和其它法庭科学结合起来”(p. 137)。使用似然率方法才是法庭比较证据评价的现代思想。我们还是鼓励世界范围内的所有法庭语音比较研究人员和检验人员尽快采用量化的似然率方法为标准。

Rose, P. (2007) Going and getting it - Foresnic speaker recognition from the perspective of a traditional practitioner-researcher. Paper presented at the Australian Research Council Network in Human Communication Science Workshop: FSI not CSI – Perspectives in State-of-the-Art Forensic Speaker Recognition, Sydney, New South Wales, Australia, 6–7 December 2007.

What I am going to talk about: 1. Logical justification of Likelihood Ratios; 2. Some examples from casework; 3. Testing the LR-based approach; 4. How Traditional Forensic Speaker Recognition has participated in the revolution in evaluation of forensic identification evidence; 5. Pros and cons of the LR-based approach; 6. The most pressing research questions for the LR-based paradigm.

Hodgson, D. (2007) Speaker identifiaction – A judicial review. Paper presented at the Australian Research Council Network in Human Communication Science Workshop: FSI not CSI – Perspectives in State-of-the-Art Forensic Speaker Recognition, Sydney, New South Wales, Australia, 6–7 December 2007

This is a field where considerable scientific work is proceeding. It may be of use to those undertaking this work, particularly to those who do so with a view to producing material that can be used as evidence in legal proceedings, to have a legal perspective on this field. I’m inclined to think that evidence in terms of likelihood ratios will probably be the most useful way to go. It does seem to be the most scientific way for this kind of evidence to be given; and I think that the courts will become experienced in handling evidence in this form and in explaining it to juries - so that the evidence will be scientifically sound and will neither confuse juries nor involve giving up the idea that proof in criminal cases must be beyond reasonable doubt.

Rose, P. (2006) Catching criminals by their voice: Combining automatic and traditional methods for optimum performance in forensic speaker recognition. Successful grant application for Australian Research Council Discovery Project. Research funded: 2007–2010.
- One or more speech samples of a known voice are compared with samples of unknown origin. The unknown samples are usually of the individual alleged to have committed an offence, and the known voice belongs to the suspect, defendant or accused. The interested parties (police, court, legal counsel) are then concerned with being able to say on the basis of the evidence whether the two samples have come from the same person, and thus be able either to identify the defendant as the offender, or exonerate them. At the moment, two very different, but complementary, approaches – Traditional and Automatic – are used to do this. The aim of this project is to improve substantially on the methods used to identify criminals by their voice by combining these two approaches.
Rose, P. (2005) Forensic speaker recognition at the beginning of the twenty-first century – An overview and a demonstration. Australian Journal of Forensic Sciences, 37, 49–72.
- This short paper illustrates some of the main factors upon which forensic speaker recognition (FSR) depends. Its aim is to clarify for interested parties the most important things about the state of FSR at the turn of the century. I will do this by first outlining a typical FSR scenario, and making precise what kind of answer can be given to the question “was the incriminating speech said by the suspect?” I will then give a demonstration with real data of how that answer can be arrived at. Finally, I use this demonstration as a background to a discussion of more general topics of relevance and interest: the evolution of FSR over the last 20 years; the different types of FSR; and the different methods.