Higher Sem. in Bilingualism. Nathan Young: Developments and challenges in forensic speaker comparison

This presentation provides an overview of forensic speaker comparison in Sweden, focusing on recent advances, persistent challenges, and the potential of automation and artificial intelligence (AI).

Demand for forensic speaker comparison has increased as speech recordings have become more prevalent in criminal investigations. The use of social media and smartphones in criminal activity, expanded possibilities for lawful interception, and the growth of telephone fraud have all contributed to this development. Meanwhile, advances in digital audio analysis, acoustic measurement, and statistical methods have made forensic speaker comparison increasingly empirical and reproducible.

A central challenge is the need for reference corpora that adequately reflect contemporary linguistic diversity. Forensic experts examine features such as fundamental frequency, vowel formants, prosody, and voice quality, but the evidential significance of similarities between recordings depends not only on the degree of similarity, but also on how common those features are in an appropriate reference population. Outdated corpora may misrepresent the typicality of shared features, particularly when speakers use linguistic varieties that are poorly represented in existing datasets.

A second challenge is the continued reliance on time-consuming manual phonetic annotation and analysis. This often restricts investigations to relatively small speech samples and can contribute to case backlogs. Developments in computational linguistics and AI have the potential to automate parts of the workflow, enable more comprehensive analyses, and make forensic speaker comparison more efficient across a wider range of cases.

The presentation will consider both the opportunities and limitations of these developments, with particular attention to the interplay between linguistic expertise, representative data, and automation.

Nathan Young