Prosody-based detection of the context of backchannel responses.

摘要: ABSTRACT Current spoken dialogue systems lack positive feedback such asbackchannels, which are common in human-human conversa-tions. To develop more natural human-computer interfaces, theinvestigation of backchannel-responses indispensable. In thispaper, we propose a method for detecting the precise timing forbackchannel responses Japanese and aim at incorporating suchmethod future systems. The proposed methodis based on machine learning technique with variety prosodicfeatures. It is shownto be effectivein automatically derivingrulesfor contexts backchannels. performance ofour considerably better than previous methods. 1. INTRODUCTION Many researchers have reported that people hesitate to talk withspokendialogue due positivefeedback fromthe as backchannels, conversations [3, 6]. investigation backchannel-responsemechanisms this paper, amethodfordetecting precisetiming responsesin spo-ken systems.In method, backchannels de-tected by using only prosodic features fundamental fre-quency energy, relatively easy handle currentspeech technology. contrast existing methods, whichuse very limited number hand-made heuristics, weemploy varietyof fea-tures might relevant detection backchannelcontext. will shown our effective automati-cally deriving rules contextsof andthat it performs methods.In Section 2, review related works inJapanese conversation automatic forbackchannels. 3, describe cor-pus used study provide definition backchannels.In 4, conduct psychological experiment order tocategorize negative whichare average humans. 5, obtain, us-ing decision tree cues best dis-criminate InSection 6, summarize paper.