电力自主移动机器人运检AI助手语音多路信号端点切分
Power Autonomous Mobile Robot Operation and Inspection AI Assistant Speech Multiplex Signal Endpoint Segmentation
杨洋 1宋祉霖 1豆朝宗1
作者信息
- 1. 中国核电工程有限公司,北京 100840
- 折叠
摘要
研究了电力自主移动机器人运检AI助手语音多路信号端点切分仿真,解决语音端点切分运算复杂、不准确问题.机器人运检AI助手语音服务包括语音多路信号采集、播报、理解、控制等,采集到的AI助手语音多路信号经客户端降噪处理后,通过主分量分析筛选语音多路信号矩阵的特征值和特征向量,确定最有用因素作为基底向量,获取语音多路信号中最典型特征,经线性区分分析使语音多路信号的特征布局集中,获取用于切分的特征并构建变化矩阵,利用MLLT计算变化矩阵,完成语音多路信号样本的协方差矩阵对角化,并将特征信息作为服务端卷积神经网络预测模型输入,从帧级上实时分类语音多路信号数据,实现语音多路信号端点的切分.仿真实验结果显示:该方法可有效去除语音多路信号噪声,并完成语音多路信号端点准确切分.
Abstract
This paper studies the simulation of voice multi-channel signal endpoint segmentation of the AI assistant of electric power autonomous mobile robot operation inspection,and solves the problem of complex and inaccurate voice end-point segmentation operation.Robot operation inspection AI assistant voice service includes voice multi-channel signal acqui-sition,broadcasting,understanding,control,etc.after the collected AI assistant voice multi-channel signal is denoised by the client,the eigenvalues and eigenvectors of the voice multi-channel signal matrix are screened through principal component analysis,and the most useful factors are determined as the base vector to obtain the most typical features in the voice multi-channel signal.After linear discrimination analysis,the feature layout of the voice multi-channel signal is centralized,Obtain the features used for segmentation and construct the transformation matrix.Use milt to calculate the transformation matrix,complete the diagonalization of the covariance matrix of the speech multi-channel signal samples,and input the feature infor-mation as the convolution neural network prediction model of the server,classify the speech multi-channel signal data in real time from the frame level,and realize the segmentation of the speech multi-channel signal endpoint.The simulation results show that this method can effectively remove the noise of speech multi-channel signals and complete the accurate segmenta-tion of speech multi-channel signal endpoints.
关键词
语音多路信号/机器人/AI助手/语音特征提取/预测模型/端点切分Key words
voice multi-channel signal/robot/AI assistant/speech feature extraction/prediction model/endpoint seg-mentation引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据