Speech Recognition Based On Neural Network And Construction Of Knowledge Map

With the progress of society and the development of science and technology,intelligent products have been widely used in all aspects of life,and speech recognition is the most natural way to realize the communication between human and machine,which greatly affects the living habits of human beings.The power grid dispatching is the key to the safe and stable operation of the power grid.The existing power grid dispatching involves a lot of repetitive human operations.With the expansion of the scale,the existing methods are not only inefficient,but also increase the risk of human error.Speech recognition is applied to the field of power grid,combined with power grid dispatching,so that automation can replace human operation.At present,there is no special thesaurus and related speech recognition model in the field of power grid.Due to the strong domain of dispatching terms,lines,faults and other words,the establishment of language model for dispatching voice in the field of power grid can improve the recognition efficiency and accuracy.Based on BRNN and dispatching voice,this paper studies the method of speech recognition.Firstly,it introduces the relevant technical principles of speech recognition,then discusses the construction of traditional acoustic model and its training methods,and then introduces the basic theory of bi-directional recurrent neural network in detail,and builds a speech recognition model in the field of power grid based on neural network,Starting from the preprocessing of voice data,the process includes pre emphasis,framing and windowing,endpoint detection,then MFCC feature extraction,and finally the construction of network model.The word error rate algorithm is used as the evaluation index of the model.For the traditional RNN based or LSTM based network structure,the current output is predicted according to the past state.The experiment uses the network with two RNNs Compared with the experimental results,the acoustic model trained by brnn structure can achieve better recognition effect.Based on the result of speech recognition,the knowledge map of power grid is constructed.By detecting the text,it can judge whether the dispatching instructions issued by the dispatcher are accurate.At the same time,knowledge map technology is applied to identify named entities by using CRF++ technology.The relationship between entities is extracted based on BiGRU-ATT.Finally,the entity and entity relationship are stored in the neo4j graph database to retrieve the possible causes of the failure,Recommend solutions to reduce search complexity.