Research Of Spoken Language Understanding Methods Based On Deep Neural Network

The human-machine dialogue system refers to a system in which a machine can talk with a user in real-time through voice,and can provide users with help or services.The system usually consists of a speech recognition module,a sp oken language understanding module,a dialogue management module,a text generation module and a text to speech module.The spoken language understanding module is mainly used to extract the domain information,intent info rmation and semantic slot information in spoken text.The spoken language understanding module plays a vital role in the human-machine dialogue system.Therefore,this paper combined a deep neural network to conduct in-depth research on spoken language understanding and applied the research results to the speech ordering dialogue system.First,the spoken text is short in length and sparse in features,and it is di fficult to improve the accuracy of intent classification.Aiming at this problem,an model of classification for Chinese short texts is proposed based on local semantic features and context relationships.In this model,the convolutional neural network is utilized to extract the local semantic features of text;whereas the bidirectional long short-term memory neural network is used to extract the semantic features of text,and the attention mechanism is combined to enable the model to extract the features most relevant to the current task from the numerous features,so as to better intent classification.Second,in the spoken language understanding task,there is a certain connection between the intent determination task and the semantic slot filling task.Due to this condition,this paper proposes a joint model for intention determination and semantic slot filling in the field of order,in which ALBERT pre-trained models and convolutional neural networks are used to complete the intent determination task,ALBERT pre-trained models and conditional random fields are used to complete the semantic slot filling task.Finally,Based on the above research results and the practical application scenario of ordering meals,this paper designs a voice ordering dialogue system that can interact with voices.The system is connected to the We Chat public account platform,and tests and statistical recognition rates of different users and different speech speeds show that the system can meet the requirements of voice order interaction.To improve the accuracy of the system's spoken language understanding,the system also adds an online learning mode.This mode is used to collect the wrong corpus,train the model and further improve the accuracy of spoken language understanding.