Forensic Phonetic and Sociolinguistics

Paul Foulkes, University of York
Malcah Yaeger-Dror, University of Arizona
Reva Schwartz, US Secret Service

This workshop will provide a tutorial and discussion on the problems and practices of forensic phonetics and sociolinguistics. The speakers will draw on real case materials and outline research findings, to draw attention to the inherent problems in forensic speech analysis, and encourage other researchers to contribute to four research domains.

1. Deciphering the Content of Difficult Recordings
Tape recordings containing information relevant to court hearings can be difficult to decipher, since speakers may have non-standard . perhaps foreign - patterns of pronunciation. However, phonetic and sociolinguistic analysis may help determine what was said.

2. Speaker Profiling
A combination of phonetic and sociolinguistic analysis of a voice can aid in establishing information about the speaker's background. The strength of conclusions varies and is highly dependent on the length and quality of material, the degree of any disguise involved, and the extent of descriptive dialectological and sociolinguistic information available.

3. Speaker Identification (SID)
As much as 80% of all forensic cases involve SID: that is, identifying, by means of comparative phonetic testing, a person heard speaking in a criminal recording. Generally, a forensic phonetician compares the voice in the criminal recording with a speech sample from the suspect, to assess the likelihood of their being the same person, as Labov (1988) did with the presumed PanAm bomber. Speech features may also vary markedly from situation to situation: with stress, conscious disguise, or the effect of speaking on a telephone added to the situational factors sociolinguists commonly expect. Different voices may also be affected in different ways. For instance, most people, but not all, speak more loudly when using a telephone, which results in a rise in average fundamental frequency. Nowadays SID usually includes both auditory and acoustic phonetic analysis. Much research focuses on determining which parameters of a voice are the most useful in identifying an individual speaker. Conclusions are dependent on reference information, including . where available - population statistics, from fields such as phonetics, phonology, dialectology, sociolinguistics, and speech pathology. Automated techniques are also under development, and will be discussed.

4. Speaker Identification by Lay Witnesses
In some cases, the perpetrator may simply have been heard by a lay witness who must be able to demonstrate recall of a voice in a voice parade or line-up. This consists of a set of recorded voices, including that of the suspect. Forensic phoneticians may be called upon either to construct the parade, or to analyze the contents of the parade to ensure that no samples stand out unfairly in relation to the others, for example through a markedly different accent, voice quality or recording fidelity. There is at present, however, no incontrovertible method of constructing or administering a voice parade. Moreover, evidence will be reviewed that shows that while recall of familiar voices is better than that of unfamiliar ones, close friends and family members can show a surprising degree of inability to recognize even the voices of personal intimates.