项目名称: 基于EMD的复杂声学环境下语音检测与增强
项目编号: No.60803087
项目类型: 青年科学基金项目
立项/批准年度: 2009
项目学科: 数理科学和化学
项目作者: 申丽然
作者单位: 哈尔滨工程大学
项目金额: 19万元
中文摘要: 话带信号是以语音信号为主,夹杂各种噪声,非线性、非平稳信号。在话带信号中有效的提取语音并对其进行增强会使语音的编码、传输等更加有效并能减少信道的负载。这些都是目前迅猛发展的通信系统(民、军用)所急切需要解决的问题。以往对话带信号的分析都是建立在富丽叶变换基础之上,因此这些分析方法必然受到富氏变换的局限。经验模态分解(EMD)的出现会给话带语音信号的分析注入新的血液。EMD 是近几年刚刚发展起来的一种全新的非线性、非平稳时间序列分析方法。本项目主要研究内容如下:1)半监督回归支持向量机函数估计的方法进行曲线拟合,从中得到更为准确的信号包络,并且能进行预测估计解决端点效应问题; 2)采用信息变差从理论上理论给出模态分离结束依据。采用自适应尺度搜索的方法进行经验模态分解从一定程度上解决模态混叠问题。3)EMD 和TEO 算子相结合进行语音信号检测;4)EMD 结合子空间理论和人耳听觉特性进行语音增强。
中文关键词: 经验模态分解;非线性非平稳信号处理;语音检测;语音增强
英文摘要: The voice band signal is nonlinear and nonstationary signal, which main component is speech signal and mingles with various noise. Exctracting effectively speech signals from voice band signal and then enhancing them is not only make the code and transmit more efficient but also can reduce the channel load. These all are needed to be solved for communication system (civil, military). Previously all the methods which were used to analysis voice band signal were based on fourier transform. So all the methods must be limited by fourier transform. EMD is a novel method to analysize nonlinear and nonstationary signal. And the method can provide a new way for voice band signal.the main researchs in this project as follows: 1)semi-supervised regression Support Vector Machine will be used to curve fitting. And then get more exactly signal envelope to solve the problem of endpoint.2) Give the rule for the mode separating. Using Adaptive scales searching method to separate the signal.3)Using EMD combining with TEO operator to detection speech signal. 4)Using EMD combining with subspace theory and Human Auditory to enhancing speech signal.
英文关键词: EMD ;nonlinear nonstationary signal processing; speech detection ;speech enhancement