Communication of trustworthiness in human and synthesised speech

This thesis investigates how trustworthiness is communicated in human and synthesised speech. Prior studies, largely based on young, white, Western participants, have overlooked how vocal trustworthiness perceptions may vary across age and ethnicity. This work addresses that gap by including under-represented groups such as older black and south Asian speakers and listeners. The thesis adopts a multi-stage design, beginning with a systematic review that maps conceptual and methodological gaps in the literature. The following studies aim to address these limitations. Study 1 introduces an open-access dataset of 1,152 speech samples from 96 demographically diverse speakers (by age, sex, and ethnicity), examining how trustworthy vs neutral vocal intent is expressed. Drawing on this dataset, Study 2 explores how perceived trustworthiness aligns with core social perceptions of warmth and competence, as shaped by vocal cues. Multi-part Study 3 extends the analysis to the role of cognitive biases, demographic variation, and trust predispositions. Finally, Study 4 compares human and real-world synthesised voices, bridging trustworthiness perception across speaker identities. Collectively, the findings converge on a set of acoustic features — perceived pitch, speech rate, HNR, shimmer and LTAS — that reliably shape trustworthiness impressions across both human and synthesised voices. Trustworthiness impressions were boosted with explicit vocal intent, especially where speaker group membership was uncertain or ambiguous. Listener predispositions shaped evaluations in distinct ways, highlighting that trustworthiness impressions are not only about how a voice sounds, but also who is listening and what they bring to this process. By integrating speaker intent, listener bias, and speaker nature (human vs synthesised), the thesis offers invaluable cross-validation of trust-relevant acoustic cues across production and perception replication of trustworthy human and synthesised voices. It advances theoretical models of vocal impression formation and offers practical guidance for designing socially attuned, trust-enhancing voice technologies for diverse user groups.