Grid Intelligent Dispatching Speech Recognition Interaction Research And Application Of System

With the continuous advancement of industrialization and the increasing tasks of national grid power dispatching,the traditional power dispatching mode can no longer meet the growing demand of grid automation dispatching.Meanwhile,with the development of artificial intelligence,human-computer interaction by means of speech recognition has become a popular research direction.In the field of power dispatching,the existing speech recognition interaction applications,such as Xunfei Xiaodu and Xiao Ai classmates,cannot effectively meet its needs for the following reasons:(1)insufficient professionalism,which leads to low accuracy rate of professional vocabulary due to insensitivity to power professional vocabulary in power dispatching recognition scenarios;(2)high cost of use,which requires building a complete corpus of power professionals and the model needs(3)limited application scenarios,and it is difficult to take into account a variety of power dispatch recognition scenarios;(4)low security,and it requires remote invocation of external network interfaces;(5)low portability,and it is difficult to adapt to local power grid dispatching systems.Therefore,in order to solve the above problems,this thesis aims to design an offline speech recognition interaction system for the professional field of electric power dispatching,which can be migrated to other professional fields to realize the speech recognition interaction needs for a special and professional field.It mainly contains the following features.(1)A differential-frequency priority language model is designed for the problems of specialization and usage cost.The differential frequency priority decoding method used in the differential frequency priority language model does not require a corpus for training,and the decoding process can focus on the recall of power specialized words in the speech annotation sequence.(2)To address the problem of limited applicable scenarios,based on the differential frequency priority model,the Pre Transformer model and the linguistic-semantic integration model are designed to target different recognition scenarios,respectively,and are integrated into the system.The former uses secondary decoding for language decoding,and performs well in the power dispatch call recognition task with large vocabulary and certain requirements on recognition rate,and its decoding WER is 2.89%;the latter determines whether the output of the language model is correct according to the power dispatch grammar rules and makes corresponding reverse correction,and in the power dispatch command recognition task with small vocabulary and fixed utterance structure,the WER is 1.03%.(3)Combined with the specific environment of Guangzhou power grid,we completed the engineering construction and deployment of this offline speech recognition interactive system,and realized the functions of four modules: speech acquisition,speech broadcast,dispatching speech recognition and message communication.The actual test results show that this system can well meet the needs of speech recognition interaction in power dispatching professional field and solve the problems of the existing system to a certain extent.It also passed the relevant acceptance appraisal and was connected to the intelligent dispatching platform of Guangzhou Power Grid.