To view this page ensure that Adobe Flash Player version 11.1.0 or greater is installed.
Feature Forensic linguistics major drugs importers, people traffickers or terrorists in cars, flats and business premises that have been bugged by the police or security service. The ubiquity of mobile telephones has resulted in many violent crimes being recorded. Victims or bystander witnesses to robberies, rapes and murders often dial 999 and the voice of the criminal is recorded over the open line in the background of the call. In nearly all other countries the criminal recordings are very often telephone intercepts (‘wire taps’). The UK, while allowing call interception to take place on a warrant from the Home Office, currently restricts the use of those recordings to investigative purposes (Regulation of Investigatory Powers Act, 2000). As they cannot be used in evidence, they are not submitted for speaker comparison tests. The reference recordings with which the criminal voices are compared are in most cases those of police interviews with the suspects. Until the mid- 1980s, someone suspected of being ‘the voice’ in a criminal recording would be asked to provide a voluntary voice sample. Needless to say, many did not. However, the enactment of the Police and Criminal Evidence Act (1984) required all police interviews to be recorded. This resulted in suspects ‘automatically’ providing voice samples – simply as a by-product of the interview procedure. Almost overnight, the volume of forensic speaker comparison work rose exponentially. So how is it done? The most prevalent method comprises a combination of two types of tests: auditory-phonetic (analytic listening) and acoustic (computer-based). The auditory- phonetic tests involve intensive, repeated listening and draw heavily upon the skills and ear training gained from a university education in phonetics. One ‘deconstructs’ the speech and listens selectively and individually to its component parts. For example, the voice quality (‘timbre’) is examined in accordance with categories encoded in the Vocal Profile Analysis Scheme developed by Professor John Laver and his colleagues at the University of Edinburgh. This involves checking the comparability of the criminal recording and the suspect’s recording against 38 different voice quality settings and assigning them a score on each. Some of the settings relate to activity within the larynx (e.g. creaky voice, breathy voice), some to muscular tension (e.g. tense/lax vocal tract) and others to supralaryngeal settings (e.g. tongue orientation, nasality). One also listens to intonation – the rise and fall of the pitch of the voice across utterances, to speech rhythm and tempo (the latter may be measured and averaged in terms of syllables per second), and to the pronunciation of consonant and vowel sounds; here one has the assistance of the International Phonetic Alphabet, an extended system of symbols and diacritical marks that enable the analyst to capture the fine grained nuances of pronunciation. The acoustic tests are carried out using specialised sound analysis software. They include examinations of voice pitch. This is measured as fundamental frequency, which equates to the rate of vocal cord vibration. Other acoustic tests involve examining the resonance characteristics of consonants and vowels using sound spectrograms. Here the analyst can visualise the speech on a computer screen and take detailed measurements from the acoustic signal. The process of fully analysing and comparing one sample against another may take in the region of 15 hours, and many people will no doubt be surprised to find that it is not nowadays done wholly automatically and with instantaneous results. Automatic speaker recognition (ASR) software is available and The auditory-phonetic tests involve intensive, repeated listening and draw heavily upon the skills and ear training gained from a university education in phonetics. One ‘deconstructs’ the speech and listens selectively and individually to its component parts. its accuracy and reliability are very good, though not perfect, when processing high quality recordings. The recordings that arise in forensic cases, having poorer sound quality (television or music in the background, people distant from the microphones, people speaking simultaneously, etc), pose particular problems for ASR. Our view is that it may be used in cases where the recordings are of sufficient quality, but as an addition to the types of testing described above, rather than as a stand-alone replacement. We are not quite up with CSI yet! The way in which the conclusions of speaker comparisons should be expressed is a hotly debated matter, with alternative frameworks being proposed and defended by different ‘camps’. However, experts are unanimous in the view that the evidence should never be seen as definitive, i.e. a criminal trial should not be ► Babel The Language Magazine | November 2012 13