Geoffrey Stewart Morrison -
geoff-morrison@forensic-voice-comparison.net
Director, Forensic Voice Comparison Laboratory
School of Electrical Engineering and Telecommunications
University of New South Wales, Sydney, Australia
Popular version of papers 1eID1, 3aSC6, 3pSC4
Presented Monday and Wednesday,
November 15 and 17, 2010
2nd Pan-American/Iberian Meeting on Acoustics, Cancun, Mexico
Dr Geoffrey Stewart Morrison is a Canadian researcher currently working in Sydney, Australia, where he is the Director of the Forensic Voice Comparison Laboratory at the University of New South Wales. He is also the organizer of the Tutorial and Special Session on Forensic Voice Comparison and Forensic Acoustics at the 2nd Pan-American/Iberian Meeting on Acoustics http://cancun2010.forensic-voice-comparison.net, and he says there is a revolution happening in forensic voice comparison.
What is forensic voice comparison?
Well, a typical scenario is that the police have an audio recording of an offender from a telephone intercept and another audio recording from an interview with a suspect. What the court wants to decide is whether the speaker on the two recordings is the same person or two different people. The task of the forensic scientist is to analyze the acoustic properties of the voices on the recordings and on the basis of that analysis present a weight-of-evidence statement to help the court to make its decision.
What is the nature of the revolution in forensic science?
It’s a revolution towards the use of the logically correct framework for the evaluation of forensic evidence and the testing of the validity and reliability of the systems and procedures used to analyze the evidence, and importantly to do the testing under conditions which reflect the conditions of the case at trial. These ideas aren’t new; arguments in favor of logically correct evaluation of evidence were made in relation to the trial of Alfred Dreyfus over a century ago.
The real breakthrough in practice came in the mid 1990s in forensic DNA comparison. Because it was new, lawyers challenged it in court, and so there was immense legal pressure to get it right. Also, because forensic DNA comparison was developed by people with a strong background in modern science they were better able to understand the arguments made by forensic statisticians about how to do it right.
In 2005 Michael J. Saks and Jonathan J. Koehler published an article in Science http://dx.doi.org/10.1126/science.1111565 in which they described the revolution, the paradigm shift, which had already occurred in DNA and predicted that it would also occur in other branches of forensic science. In 2009 the National Research Council Report on Strengthening Forensic Science in the United States http://www.nap.edu/catalog.php?record_id=12589 called for support for research and development so that other branches of forensic science could emulate DNA.
I don’t think either Saks and Koehler or the NRC report fully understood what it revolutionary about forensic DNA practice: they were very concerned about validity and reliability, but they said nothing about the logically correct framework. In an article in Science & Justice in 2009 http://dx.doi.org/10.1016/j.scijus.2009.09.002 and again in my introduction to forensic voice comparison recently published in the Expert Evidence series http://expert-evidence.forensic-voice-comparison.net I argued that it is the combination of measuring validity and reliability and the use of the logically correct framework which constitute the new paradigm, and this is what should be adopted in other branches of forensic science including forensic voice comparison.
This stuff contributes to decisions about whether to lock someone up or set them free. The stakes are very high. It’s important that we as forensic scientists do our jobs right.
What is the logically correct framework for the evaluation of forensic evidence?
It’s actually very simple.
Say you measure the acoustic properties of the audio recordings of the offender and the suspect and you find them to be very similar; does this mean that the two recordings must have been produced by the same speaker? Not at all. If the acoustic properties of the voices on the two recordings were very typical then you would easily get equally or more similar recordings just by randomly picking recordings from any two people in the population.
On the other hand, if the acoustic properties are relatively atypical then randomly picking recordings from any two people in the population is likely to result in less similarity.
What has to be assessed is not just the similarity of the voices on the two recordings but also their typicality. What the forensic scientist should present in court as a strength-of-evidence statement is the ratio of the similarity to typicality; this is called the likelihood ratio.
Usually the prosecution says the two recordings are of the same speaker and the defense says they are of different speakers. The likelihood ratio is a calculation of the probability of obtaining the evidence, the acoustic properties of the two voice recordings, under the same-speaker hypothesis versus under the different-speaker hypothesis.
I argue that both the similarity and the typicality should be assessed in as objective a manner as possible, this means making acoustic measurements of both the suspect and offender recordings and also of a large number of speakers which constitute a representative sample of the relevant population of speakers. Statistical models then have to be used to calculate the likelihood ratio.
Tell us about the Tutorial and Special Session you’re organizing.
When most people think of forensic voice comparison, they think of what they have seen on TV: the two audio recordings get turned into pictures then the forensic scientist looks at the pictures and says they are a match, they were produced by the same speaker. This isn’t objective, isn’t consistent with the logically correct framework for the evaluation of evidence, and its validity and reliability aren’t tested under conditions relevant to the case at trial. It reflects an approach which was popular in the 1960s through 80s which was very controversial and which has been widely discredited. The aftermath of this was that research in forensic voice comparison was largely abandoned in the United States and in the Americas in general.
Meanwhile, since the mid 90s a small number of researchers in other parts of the world have begun research on forensic voice comparison conducted within the new paradigm. There now seems to be increasing interest in forensic voice comparison in the Americas, and I thought that the Pan-American/Iberian Meeting on Acoustics would be the ideal venue to bring together some of the leading names from around the world and promote research in the Americas on forensic voice comparison conducted within the new paradigm. Since the Acoustical Society of America includes both signal-processing engineers and phoneticians it is also an ideal venue to bring together researchers who use so-called automatic and acoustic-phonetic approaches to forensic voice comparison.
We have some of the pioneers of applying the new paradigm to forensic voice comparison presenting in the Special Session.
Didier Meuwly and Andrzej Drygajlo did pioneering work on automatic approaches in the late 90s in Lausanne, Switzerland. In addition to the technical work of the Swiss group, in 2000 Dr Meuwly along with Christophe Champod published an influential paper in Speech Communication http://dx.doi.org/10.1016/S0167-6393(99)00078-3 describing the logically correct framework as applied to DNA and arguing that it should be applied to forensic voice comparison. Dr Drygajlo, Head of the Speech Processing and Biometrics Group at the Swiss Federal Institute of Technology http://people.epfl.ch/andrzej.drygajlo, will be talking about the evaluation and interpretation of forensic voice comparison evidence using automatic approaches, and Dr Meuwly, now Principal Scientist at the Netherlands Forensic Institute http://www.nederlandsforensischinstituut.nl, will be talking about improvements made in automatic approaches over the last 12 years.
Philip Rose did pioneering work on acoustic-phonetic approaches in the early 2000s in Canberra, Australia. His 2002 book Forensic Speaker Identification http://www.taylorandfrancis.com/books/details/9780415271820 has become a classic in the field. Dr Rose, Reader Emeritus at the Australian National University, will be presenting research on combining acoustic-phonetic and automatic approaches.
Another pioneering group working on automatic approaches since the early 2000s is based in Madrid, Spain. Dr Daniel Ramos, Assistant Professor in the Biometric Recognition Group at the Autonomous University of Madrid http://arantxa.ii.uam.es/~dramo, will be presenting research on dealing with an important problem in forensic voice comparison: differences introduced into audio recordings because of differences in speaking style, acoustic environment, and recording and transmission equipment. An audio recording of a person having a lively mobile telephone conversation outdoors recorded from a telephone intercept may sound very different from a recording of the same person responding in a subdued manner to questions asked in a police interview room recorded directly using a microphone.
We also have a number of other researchers from Austria, Canada, China, Germany, Spain, the UK, and the US who will be talking on these and other topics related to forensic voice comparison and forensic acoustics. In other sessions there are also researchers from Chile and Mexico presenting related work.
The whole thing is going to be kicked off by the Tutorial on Forensic Voice Comparison and Forensic Acoustics which Dr Ramos and I are giving on the Monday evening.
Apart from attending the Tutorial and Special Session at the Pan-American/Iberian Meeting on Acoustics, where can people find out more about forensic voice comparison?
I would recommend my introduction to forensic voice comparison in the Expert Evidence series http://expert-evidence.forensic-voice-comparison.net as a starting point. It’s designed to introduce the field in a relatively non-technical way which can be understood by readers without a background in forensic science, phonetics, signal processing, or statistics – the idea is that it would help lawyers, judges, and jury members to understand the testimony of forensic-voice-comparison experts when it is presented in court.
Dr. Morrison’s research is funded/supported by the Australian Research Council, the Australian Federal Police, the National Institute of Forensic Science Australia, the Australasian Speech Science and Technology Association, the Western Australia Police, the New South Wales Police, the Civil Guard of Spain, the Ministry of Science and Innovation of Spain, the United States Government Office of the Director of National Intelligence (Intelligence Advanced Research Projects Activity through the Army Research Laboratory), and the International Association for Forensic Phonetics and Acoustics. All statements of fact, opinion or conclusions expressed by Dr Morrison are his personal belief and should not be construed as representing the official views or policies of any of the funding/supporting organizations.
http://geoff-morrison.net
http://forensic-voice-comparison.net
http://forensic.unsw.edu.au