Research On Key Technologies Of Intelligent Speech Interaction Based On Graph Structure

With the advancement of science and technology and society and the rapid development of artificial intelligence technology,human-computer interaction no longer depends solely on the keyboard,mouse,or display.Because human-computer interaction through voice is more in line with human communication habits At the same time,it is more convenient,so controlling the behavior of computer systems through voice is more and more popular.Such systems are generally based on artificial intelligence technology,so they are called intelligent voice interaction systems.Intelligent voice interaction and related technologies are computers.An important branch of the future development of the system.Intelligent voice interaction systems are currently divided into two types: chat and task.The task-based interaction system aims to complete specific tasks.It aims to receive instructions from the user to the computer in the form of voice,and then analyzes the user's real needs by means of a series of algorithms.After making an effective response,the response form of the intelligent voice interaction system is not limited to voice,but can also be in various forms such as the execution result of an action and the update of the interface.The purpose of this paper is to study the key technologies of intelligent speech interaction systems.The main work is focused on the analysis and design of the semantic understanding module and the session management module.First,by analyzing the voice interaction process and related content in the automotive field,and then summarizing its characteristics,then,through multiple sets of comparative experiments,a solution that can meet the needs is obtained.Finally,the system is designed and designed.achieve.In this system,the core function of the semantic understanding module is the extraction of key information.This paper proposes the concept of word slot order.Using the word slot order model can accurately extract the system's interesting content from the user input short sentences,but the word slot order The essence of the model is a templatebased method.It has a certain generalization ability,but it is not very strong.After experiments,it is found that the neural network model can make up for the above shortcomings.Slot-sequence and Bi-LSTM-CRF network hybrid models are used to perform key information extraction tasks.The core function of the session management module is intent recognition and management of conversation context.For the intent recognition task,this paper proposes a hybrid model of word slot order and TEXTCNN network.This model can alleviate to a certain extent the lack of high-quality training data sets in the research area of this article.The problem.Aiming at the task of session context management,this paper proposes a rule-based solution,which enables the system to have certain multi-round interaction capabilities.