Interpretation of user evaluation for emotional speech synthesis system

Abstract

Whether it is for human-robot interaction or for human-computer interaction, there is a growing need for an emotional speech synthesis system that can provide the required information in a more natural and effective manner. In order to identify and understand the characteristics of basic emotions and their effects, we propose a series of user evaluation experiments on an emotional prosody modification system that can express either perceivable or slightly exaggerated emotions classified into anger, joy, and sadness as an independent module for a general purpose speech synthesis system. In this paper, we propose two experiments to evaluate the emotional prosody modification module according to different types of the initial input speech. And we also provide a supplementary experiment to understand the apparently prosody-independent emotion, or joy, by replacing the resynthesized joy speech information with original human voice recorded in the emotional state of joy.