Enhancing Expressiveness of Synthesized Speech in Human-Robot Interaction: An Exploration of Voice Conversion-Based Methods

The utilization of speech synthesis systems, particularly text-to-speech (TTS), in human-robot interaction poses a persistent challenge. Users frequently perceive synthesized speech as lacking natural prosody variation, which can impede the effectiveness of such interactions. In response to this challenge, this paper delves into the realm of voice conversion-based methods, with the aim of imbuing synthesized speech with enhanced expressiveness by extracting prosody features from users. A comprehensive investigation is carried out, encompassing the implementation of various established and widely used techniques. To ensure the relevance and practicality of the study, content derived from hospital patient education materials is employed as the primary source material. The outcomes of the experiments reveal that existing voice conversion (VC) and emotional voice conversion (EVC) methods do not substantially enhance the naturalness of synthesized speech. Moreover, a discovery is made that evaluation using quantitative methods that are common in voice conversion individually does not necessarily reflect the subjective experience of synthesized speech. These findings provide a step forward in advancing the field of speech synthesis and human-robot interaction, offering valuable insights for future research and development in this area.