Behavioral Effects of AI-generated Synthetic Vocalizations in Free-Moving Marmosets

Woolgar, E; Du, Y; Errington, S; Slater, B; Ogawa, T; Kikuchi, Y

doi:10.1109/CyberSciTech68397.2025.00122

Behavioral Effects of AI-generated Synthetic Vocalizations in Free-Moving Marmosets

Lookup NU author(s): Emma Woolgar, Steven Errington ORCiD, Dr Ben Slater ORCiD, Dr Yuki Kikuchi ORCiD

Downloads

Full text for this publication is not currently held within this repository. Alternative links are provided below where available.

Abstract

AI-based voice synthesis technologies are central to improving human-computer interaction (HCI). Socially assistive robots, which incorporate such voice technology, are emerging as promising tools for enhancing the quality of care for individuals with cognitive decline. However, the development of emotionally expressive artificial voices, particularly for older adults, remains limited. A key challenge is the lack of understanding of how auditory and emotional cues in artificial voices influence perception and behavior in cognitively impaired populations. To address this, it is important to study how deep learning models used for voice synthesis affect behavior in animal models that can translate to human clinical conditions.In this study, we used a Variational Autoencoder (VAE) to generate synthetic vocalizations and compared the behavior of three freely moving marmosets in response to AI-generated and natural vocalizations. We observed a significant main effect of condition on two behaviors (stationary and leg stand - contact), suggesting that vocalization type influenced behavioral responses. However, post hoc pairwise comparisons did not show statistically significant differences between specific conditions, possibly due to limited power or individual variability, suggesting that AI-generated vocalizations were not perceived as equivalent to natural ones. These findings highlight the need for more diverse training datasets to develop VAE models for this species. Despite current limitations, marmosets remain a promising model for advancing synthetic voice technologies to enhance communication and cognitive outcomes in humans.