How context shapes the appropriateness of a robot’s voice

Abstract

Social robots have a recognizable physical appearance, a distinct voice, and interact with users in specific contexts. Previous research has suggested a 'matching hypothesis', which seeks to rationalise how people judge a robot's appropriateness for a task by its appearance. Other research has extended this to cover combinations of robot voices and appearances. In this paper, we examine the missing connection between robot voice, robot appearance, and deployment context. In so doing, we asked participants to match a robot image to a voice within a defined interaction context. We selected widely available social robots, identified task contexts they are used in, and manipulated the voices in terms of gender, naturalness, and accent. We found that the task context mediates the 'matching hypothesis'. People consistently selected a robot based on a vocal feature for a certain context, and a different robot based on the same vocal feature for another context. We suggest that robot voice design should take advantage of current technology that enables the creation and tuning of custom voices. They are a flexible tool to increase perception of appropriateness, which has a positive influence on Human-Robot Interaction. I. INTRODUCTION Spoken communication is the primary form of interaction between a growing number of social robots and their users. However, the overall effort of designing robot voices is considerably less than the amount of work that goes into designing their physical appearance [1]-[4]. Research in Human-Robot Interaction (HRI) could exploit the flexibility afforded by many available Text-to-Speech systems, voice banks, and custom recordings, enabling designers to gain greater control over how robots are perceived. Voices contribute to impression formation of newly-met individuals in human-human interactions [5], as well as shaping how impressions develop over time [6]. Besides linguistic content, voices carry a wide variety of information, ranging from indexical characteristics of the speaker such as gender, age, and place of origin, as well as temporary state alterations, like mood, emotions, or health [7], [8].

Use of a humanoid robot for auditory psychophysical testing

“…The auditory performance results of the present study show promise for conducting the PICKA psychophysics tasks, insofar as the voice cue sensitivity and voice gender categorisation tests are concerned on Sam. Furthermore, our results contribute to the growing potential of using humanoid robots for both learning and testing applications [34-36, 75, 76], also for specific target groups and rehabilitation applications (e.g., [43,77,78]), and our general understanding of speech communication and voice perception in HRI, an increasingly relevant topic for social robots [1,13,79,80].…”

Is Natural Necessary? Human Voice versus Synthetic Voice for Intelligent Virtual Agents

“…The impact of matching the agent's look with its voice (congruency) on user-agent trust has been further studied, with contrary findings. For example, Torre et al [44] introduced an investment game to 120 participants to play with either a generous or a mean robot that switches from a human to a synthetic voice (or vice versa) in the middle of the game. The results indicate the importance of matching the voice with the look to form the first impression, which impacts the user's trust in the agent.…”

“…In a matching task, Torre, Latupeirissa, and McGinn [44] asked 60 participants to match one robot out of eight robots' pictures with a voice from a set of voices that varied in terms of gender and naturalness, for four different contexts: home, school, restaurant, and hospital. The results showed that the more mechanical the look of the robot, the greater the tendency for the robot to be matched with a synthetic voice, and that the matching was significantly different across the contexts.…”

Do Not Freak Me Out! The Impact of Lip Movement and Appearance on Knowledge Gain and Confidence

“…This suggests that outcomes were related to the congruence of verbal and non-verbal features. In contrast, other studies failed to prove the importance of congruency on the desired outcome [13,14], which was explained by the dependency on the context such as education [15] and entertainment [16].…”