语音图信号处理理论与技术研究  

Research on speech graph signal processing theory and technology

作　　者：杨震[1,2] 王婷婷[1,2] YANG Zhen;WANG Tingting(National Engineering Research Center of Communication and Network Technology,Nanjing University of Posts and Telecommunications,,Nanjing 210003,China;College of Telecommunications&Information Engineering,Nanjing University of Posts and Telecommunications,Nanjing 210003,China)

机构地区：[1]南京邮电大学通信与网络技术国家工程研究中心,江苏南京210003 [2]南京邮电大学通信与信息工程学院,江苏南京210003

出　　处：《南京邮电大学学报（自然科学版）》2020年第5期43-51,共9页Journal of Nanjing University of Posts and Telecommunications：Natural Science Edition

基　　金：国家自然科学基金(61671252,61271335,61901229,62071242)资助项目。

摘　　要：首先论述新型图信号处理(Graph Signal Processing,GSP)技术的基本概念和相关研究进展,其中包括图拓扑结构、图傅里叶变换、滤波及图学习。鉴于语音信号是一种非平稳和非线性的信号,为此,文中研究基于GSP技术的语音信号的图映射理论,即将时域语音信号映射为图域的语音图信号,通过设计语音图信号的图拓扑结构和图邻接矩阵来研究语音图信号的内在潜藏关系,进而设计优于经典DSP的语音图信号消噪算法和系统。在此研究基础上,进一步讨论相关实际应用问题,例如麦克风阵列环境下如何通过GSP技术处理麦克风阵列声源(说话人)的定位及追踪问题。期望基于GSP技术语音图信号处理理论,为语音识别、合成、编码、增强等各个领域的图信号处理奠定理论基础。The essential concepts and recent developments of a new graph signal processing(GSP)tech­nology are discussed,including graph topology,graph Fourier transform,filtering,and graph learning meth­ods.Since the speech signal is a non-stationary and nonlinear signal,GSP technology can be used to map speech signals into speech graph signals.Then,the hidden relationships among speech samples by defi­ning graph topology and adjacency matrices are studied to expect superior processing performances to the classic DSP when GSP is applied to the de-noising system of speech signals.On this basis,howto deal with the localization and tracking of microphone array sound source(speaker)is discussed in microphone array environment by the GSP technology.It is expected that speech signal processing on graphs will lay a theoretical foundation for subsequent speech recognition,speech synthesis,speech coding,and speech en­hancement,etc.

关 键 词：GSP 图傅里叶变换 图滤波器 语音增强 声源定位及追踪 

分 类 号：TN912.3[电子电信—通信与信息系统]