A Human-Computer Dialogue Control Method And System Implementation Based On User Registration Information

Speakers voice authentication technology is rapidly entering the daily life of people,however,because of the extreme convenience of sound recording,easily cause the risks of imitation,playback,etc,etc.The remote speaker authentication system which this paper is based on made use of channel characteristics to resist recording playback attacks.In practice,howerver,because the authentication use the traditional methods such as text repeating and random questions to collect voiceprint of users,it may cause bad effects on the pronunciation quality and the results of authentication when people is not familiar whith the text or the questions.In order to solve this problem,this paper designs a human-machine dialogue control method which is in accordance with the daily conversational habits of the registered users.Through asking and answering,the system friendly and randomly interacts with users in voice to ensure that users can successfully complete the recording process under machine's control.The main work and contributions in this paper are as follows:1.A novel dialogue manage model is proposed.It makes use of the hierarchical probabilistic topic state machine based on keyword-detection to model the process of conversation.This model can manage the multi-topic human-computer conversation effectively,and the dialogue manager implemented by this method has the advantages of high modularity,high reusability and scalability.2.A natural language question generation scheme based on RDF knowledge graph is proposed.It makes use of the diversity and the correctness of knowledge graph to ensure the diversity and rationality of the machine-generated questions.3.Drawing on the grammar of regular expressions,we design a template language to generate natural language questions which can make use of the knowledge graph and the registered user's attributes.The grammatical definition and the operational semantics are also given in this paper.4.With the methods of all above,we implement a prototype system of human-computer dialogue based on the user registration information.This system models the dialogue process with the hierarchical probabilistic topic state machine based on keyword-detection to generate dialogue management policies wich conforms to the user's characteristics,and generate natural language questions with the template language and the RDF Knowledge Graph.At the end of the paper,examples of the prototype system are given,which show that the dialogue process controlled by the system has good randomness and naturality.