The Realization And Key Technology Research Of The Voice Dialogue Robot System Based On ALICE

With the rapid development of artificial intelligence and human-computer interaction technology,voice chatbot has become the research hotspot and the main form of realization of intelligent speech dialogue system,which has extremely important practical significance to speed up the process of intellectualized human-computer interaction.What is more,it has a high application value in many fields,such as national defense military affairs,commerce,and education and so on.Hence,current chatbots based on simple template matching technology can hardly meet the increasing need from both academic and industrial community.It is an inevitable trend to develop a voice chatbot that can understand semantics and recognize voice emotion.Therefore,we make research on voice chatbot in the thesis.ALICE is a text chatbot,developed with heuristic template matching strategy,which is considered as one of the best systems among the same type of chatbot.In this paper,a voice chabot based on ALICE is designed and realized,aiming at solving the problems of single template matching mode and lack of the ability to recognize human voice emotion.We have done a series of research to these problems.The main work in this thesis is shown as follows.(1)We have studied the ALICE chatbot system in detail.A voice chatbot system based on ALICE is realized by combining speech recognition and speech synthesis technology,and a simple AIML dialogue database has been build,achieving the function of man-machine voice conversation.(2)We have studied the similarity measure methods of Chinese sentences.For the problem of single pattern matching in ALICE voice chatbot prototype system,we proposed a wighted word matching(WIT)similarity measure method.The test shows that the WIT algorithm realized the flexible matching of the template in AIML,and that improved the accuracy of ALICE voice chatbot system's understanding of users' intention.Moreover,the system achieved satisfactory performance in real-time requirement.(3)We have studied voice emotion recognition technology in depth.Based on the CASIA Chinese emotion corpus,we extracted 384 sound feathers by using openSMILE.SVM algorithm and Momentum BP-ANN algorithm were used in this thesis to classify emotion into six categories,including anger,happy and sad,and so on.The results show that the accuracy of speech emotion recognition classification based on SVM and BP-ANN is 60.33% and 63.33% respectively,which achieves a high level in Chinese speech emotion recognition at present.