摘要
为提高智能电网调度控制系统中语音识别的准确性,提出一种基于格拉姆角场(Gramian Angular Field,GAF)和深度残差网络(Deep Residual Network,DRN)的电力调度语音识别方法。首先利用GAF中的格拉姆角差场和格拉姆角和场两种方法将一维时间序列语音信号转化为二维特征图像;然后采用DRN对语音信号的二维特征图像进行特征提取和识别模型建立。采用实际电力调度语音信号对模型进行训练及测试,结果表明,该模型可有效识别电力调度语音,识别准确率超过99%。
In order to improve the accuracy of speech recognition in smart grid dispatching and control system,a power dispatching speech recognition method based on Gramian Angular Field(GAF)and Deep Residual Network(DRN)is proposed.Firstly,1-dimensional time series speech signals are transformed into 2-dimensional feature images by using two methods of Gramian Angular Difference Field and Gramian Angular Summation Field of GAF.Then,the DNR is used to extract the 2D feature image of speech signal and establish the recognition model.The model is trained and tested by using actual power dispatching voice signals,the results show that the model can recognize power dispatching speech effectively,and the recognition accuracy is more than 99%.
作者
姚永波
焦小龙
王晓波
YAO Yong-bo;JIAO Xiao-long;WANG Xiao-bo(State Grid Xinjiang Information&Telecommunication Company,Urumqi 830000,China)
出处
《信息技术》
2022年第9期169-173,179,共6页
Information Technology
关键词
电力调度
自动语音识别
格拉姆角场
深度残差网络
二维特征图像
power dispatching
automatic speech recognition
Gramian Angular Field
Deep Residual Network
two-dimensional feature image