Dialect Proximity and Soft Biometric Inference from Read Speech

In forensic phonetic casework, listeners may be required to form impressions of an unknown speaker based on speech alone, often without specialised training. Such lay speaker profiling can involve inferences about a speaker's regional background as well as other social or soft biometric characteristics. In this study, we examine how accurately lay listeners make such inferences and which listener-related factors are associated with identification performance, based on two constituent substudies. In Part I, 286 Swiss German listeners participated in an online perception task in which they estimated the age, regional origin, education level, sexual and political orientation, height, and weight of 30 Swiss German speakers reading from a script. Overall, listener performance for most factors was limited and often shaped by response tendencies and sample imbalances rather than by reliable perceptual sensitivity. However, regional origin emerged as the factor by which listener judgments were most systematically structured, particularly when evaluated using continuous, distance-based measures. Identification accuracy on the municipality level improved as the linguistic proximity between listener and speaker dialects increased, indicating that regional inference is strongly shaped by listener–speaker dialect relations.

In Part II, a small set of linguistic and acoustic features of the speakers’ read-speech productions was examined to contextualise perceptual patterns observed in Part I. While these analyses did not reveal stable correlates for most social factors, they suggest that task-specific fluency cues may have influenced some listener judgments. Overall, the findings highlight both the limits and the conditions under which lay listeners can infer social and soft biometric information from speech, with particular implications for the cautious interpretation of such judgments in forensic settings.