Toward Construction of Spoken Dialogue System that Evokes Users' Spontaneous Backchannels

This paper addresses a first step toward a spoken dialogue system that evokes user’s spontaneous backchannels. We construct an HMM-based dialogue-style text-to-speech (TTS) system that generates human-like cues that evoke users’ backchannels. A spoken dialogue system for information navigation was implemented and the TTS was evaluated in terms of evoked user backchannels. We conducted user experiments and demonstrated that the user backchannels evoked by our TTS are more informative for the system in detecting users’ feelings than those by conventional reading-style TTS.