Related Word Recommendation Mechanism for Speech Dialogue System

Abstract

In recent years, speech dialogue systems have de-veloped remarkably. This university has developed and publishedthe "MMDAgent" voice interaction system toolkit as open source software. MMDAgent uses "Julius" and "HTS" for speech recognition and speech synthesis. Its expressive agent comprises these combined with 3-D CG rendering module. As an example of MMD Agent application, "Smart Mei-chan" has been developedand transplanted to smartphones. Although "Smart Mei-chan"has extremely high performance as a voice dialogue system, considered from the viewpoint of the user, a difficulty persists that has remained unresolved: novice users have no idea whatto talk about. To induce attractive conversation, we develop arelated word recommendation mechanism for "Smart Mei-chan".Thereby, novice users will be given topics with which they canhave a good conversation. Simultaneously, we demonstrate the attractiveness of the system using recommendation words for conversation.