The Effects of Auditory Social Cues in Interactions with the BLISS Conversational Agent.
Keywords
Loading...
Authors
Issue Date
2022-07-13
Language
en
Document type
Journal Title
Journal ISSN
Volume Title
Publisher
Title
ISSN
Volume
Issue
Startpage
Endpage
DOI
Abstract
Objective: The objective of this study was to measure the effects of auditory cues from expressive voice qualities on the perceived social role, and social presence of the BLISS ‘Babbelbot’; a CA purposefully designed to improve well-being of elderly. Apart from role perception and social presence, self-perceived user engagement was measured to test whether expressive voice quality improved engagement with the interaction.
Methods: The study uses a between-subjects experiment, where one group interacted with the version of the BLISS Babbelbot that used a TTS expressive-acoustic model, which was trained using a DNN. The expressive model used an encoder embedding to inform on the phonetic input, the relations between phonemes, as well as stress and word boundary information. The control group received the same interaction, but with a non-expressive TTS model. To ensure comparability, the groups received the same prompts from the system. To measure role perception, social presence, and user engagement, a post-interaction survey was used, presenting Likert-scale statements from the User Experience Scale (O’Brien, 2018) for engagement, and the Networked Minds Measure (Biocca and Harms, 2001).
Results: The results showed few interpretable differences between the expressive (N = 20) and neutral (N = 26) groups. With regards to role perception, all participants indicated little comparability to conversations with a friend/companion, assistant, or healthcare worker. The neutral group did indicate significantly more comparability to an interaction with a friend/companion than the expressive group (t (42) = 2.03, p = .048, at p .05), as well as less frustration (t (44) = 2.58, p = .013) and distraction (t (44) = 2.16, p = .037) caused by BLISS.
Conclusion: The comparisons of the experimental group and the control group did not exhibit statistically significant differences caused by the expressive voice qualities. If anything, these voice qualities seemed to have a negative effect on perceived frustration and distraction rates during the conversation. As such, the experiment did not yield the hypothesized results, nor did provide a conclusive answer to the research questions. However, this could also be due to the several technical difficulties or challenges, as well as the relatively slow pace of both answer transcription and prompt production in the system. To circumvent technical challenges or sooth frustrations regarding the pace or functionality of the system, future studies should consider an in-person experimental setting rather than an online setting. Future studies regarding CA would also benefit from standardized reporting, to increase both the interpretability and comparability of the results. Keywords: Conversational Agents, CASA, Auditory Social Cues, Role Perception, Social Presence, User Engagemen.
Description
Citation
Supervisor
Faculty
Faculteit der Letteren