Perception of synthetic speech with emotion modelling delivered through a robot platform: an initial investigation with older listeners

Abstract

In this paper we give results of an initial investigation into the perception of synthetic speech delivered through a robotic platform. The robotic speech was judged by 19 residents and 10 staff of a New Zealand retirement village. We have investigated intelligibility and quality measures on two English language diphone voices, with US and New Zealand accents. We have also looked at the effects intonation modelling has on these measures. Our results indicate that the New Zealand voice is preferred and scores higher in the quality measure, additionally we see evidence that the dialogues delivered through both voices are intelligible. We also observe a difference in opinion to the intonation modelling. Comparing the results between staff and residents, we see that residents give lower scores to intelligibility and quality measures.