基于HMM与RBF混合模型的情感智能聊天系统  

Emotional Intelligence Chat System Based on HMM and RBF Mixture Model

作　　者：闫丹阳 姜梅 耿秀丽 闫伟 YAN Dan-yang;JIANG Mei;GENG Xiu-li;YAN Wei(School of Computer Science and Engineering,Shandong Normal University,Jinan 250000,China)

机构地区：[1]山东师范大学信息科学与工程学院,山东济南250000

出　　处：《计算机技术与发展》2018年第4期109-113,118,共6页Computer Technology and Development

基　　金：教育部留学回国基金;山东省高等学校科技计划项目(J15LN26);山东师范大学大学生创新创业训练计划项目

摘　　要：当前的智能聊天系统多采用文字进行交流,存在不能准确回答问题、转移话题、答非所问等一系列问题。因此,对情感智能聊天系统进行细致的研究、改进和升级,不仅可以推动系统的发展,还可以为人们提供更为人性、智能化的服务。文中提出了一种基于HMM与RBF的混合模型,创建人类情感语音库,以还原人类最真实的情感感受,并利用Flex技术进行情感语音库的动态更新;同时,运用语料库标注体系,以标注规范、纠错机制、补充学习作为语料库质量监控手段,从而保证语料库的完备性。在该系统下用户既可以采用文字聊天,又能进行语音聊天,并在后台产生文字聊天记录,突破了现有系统只能用文字聊天的局限性。The existing intelligence chat system mainly communicate by text,which cannot accurately answer the question,shift the topic,and obtain all relevant answers.Thus,the further study,improvement and upgrading of emotional intelligence chat system can not only promote its development,but also provide people with a more humanized,intelligent service.In this paper,we propose a HMM and RBF mixture model to build the human emotional speech libraries which is dynamically updated based on Flex technology,catching the most real emotions of human experience.At the same time,this system,which uses the corpus annotation scheme,ensures the completeness of corpus by means of a corpus quality monitoring method of the tagging criterion,error correction mechanism,supplementary learning.Users can chat with text or voice by this system which generates text chat record,breaking the existing limitations of using text chat.

关 键 词：隐马尔可夫模型 径向基函数 FLEX技术 情感智能 语料库收集 

分 类 号：TP302.1[自动化与计算机技术—计算机系统结构]