Title:
Rule-based Emotional Voice Conversion Utilizing Three-Layered Model for Dimensional Approach

Speaker:
XUE, Yawen (JAIST)

Abstract:
The purpose of this study is to propose an emotional speech synthesis system using three-layered model based on a dimensional approach. The content of three-layered model is as following: acousitc features in the bottom layer, semantic primitives in the middle layer and emotion dimensions in the top layer.  In order to estimate emotion dimensions, Fuzzy Inference System is used as a bridge to connect three layers.  Listening results from human subjects show that the improved estimated acousitc features and enhanced modification method give rise to the better-quality of synthesized emotional speech. The synthesized emotional speech can give the same impressions and similar intensity as intended.