Perceptual Effects of Ambient Sound on an Artificial Agent's Rate of Speech

Abstract

Interactive robots are increasingly being deployed in public spaces that may differ in context from moment to moment. One important aspect of this context is the soundscape of the robot and human's shared environment, such as an airport that is noisy during a weekend rush hour, yet quiet on a weekday evening. Just as humans are adept at adapting their speech appropriately to their environment, robots should adjust their speech characteristics (e.g. speech rate, volume) to their context. We studied the effect of a shared auditory soundscape on the perceived ideal speech rate of an artificial agent. We tasked raters to listen to a combination of text-to-speech (TTS) samples with different speech rates and soundscape samples from freesound.org and to evaluate the appropriateness of the speech combination and social perception of artificial speech. Contrary to our expectations, faster artificial speech in louder environments and slower speech in quieter environments were not preferred by raters. This suggests that further research into how exactly to adapt artificial speech to background noise is necessary.