基于结构相似性的语音信号增强

发布时间：2018-03-15 10:39

本文选题：非局部均值　切入点：语音增强　出处：《西安电子科技大学》2014年硕士论文　论文类型：学位论文

【摘要】：现实生活中的语音不可避免的要受到周围环境的影响,语音增强是解决噪声污染的有效方法,它的首要目标就是在接收端尽可能从带噪语音信号中提取纯净的语音信号,改善其质量。近年来,基于结构相似性的信号处理方法得到了广大学者的关注,非局部均值去噪算法作为一种利用信号结构间相似性的滤波算法,因其出色的去噪效果,成为众多学者广泛研究的对象。其通过建立相似度计算函数来求待增强点与其搜寻窗内相似点之间的权值,然后对搜寻窗内所有相似点进行加权平均得到当前点的增强结果。但是,由于语音信号噪声是非平稳一维信号,所以非局部均值滤波不能简单的直接用于语音信号的增强。针对语音信号的特点,本文提出了利用修改的非局部方法对语音信号进行增强。主要工作包含以下两个方面:(1)本文提出将非局部均值算法应用于语音增强上,且主要分为两个阶段,第一阶段对语音进行修改谱减法预处理,第二阶段使用修改非局部均值滤波对预处理后语音进行处理。最终,将非局部均值算法推广用于语音增强领域。(2)设计一种利用非局部均值算法在语音功率谱上对语音增强的算法,通过谱减法对功率谱进行预处理,得到预处理结果,结果中依然含有较大噪声,继而分析残留噪声,修改非局部均值滤波,对预处理后的功率谱进行非局部均值滤波,进一步提升了语音质量。本论文工作得到了高等学校学科创新引智计划(111计划)(No.B07048)以及教育部“长江学者和创新团队发展计划”(IRT1170资助)。
[Abstract]:In real life, the voice will be inevitably influenced by the surrounding environment, speech enhancement is an effective method to solve the noise pollution, its primary goal is at the receiving end as possible to extract pure speech signal from noisy speech signal and improve its quality. In recent years, the base has been the attention of scholars like signal processing method in the structure, non local mean denoising algorithm as a method of using signal structure similarity filter algorithm, because of its excellent denoising effect, has become the research object of many scholars. Through the establishment of the similarity function to seek to enhance its search window similarity weights, and then weighted to all the search window to get the average similarity enhancement results of the current point. However, the speech signal is non-stationary one-dimensional signal, so the non local mean filter can not be simply Directly for speech enhancement. According to the characteristic of speech signal is proposed in this paper uses a modified non local method to enhance the speech signal. The main work includes the following two aspects: (1) this paper puts forward the non local means algorithm is applied to speech enhancement, which is mainly divided into two stages, the first stage of the speech modified spectral subtraction pretreatment, second stage using a modified non local mean filter after pretreatment of speech processing. Finally, the non local means algorithm used in speech enhancement field. (2) a design using the non local mean algorithm of speech enhancement in speech power spectrum algorithm, through the power spectral subtraction spectral pretreatment, preprocessing results, results still contain large noise, and then analyze the residual noise, modified non local mean filtering of power after pretreatment of the spectrum of non local mean filter Wave, to further improve the speech quality. This thesis is supported by the higher school subject innovation engineering plan (111 plan) (No.B07048) and the Ministry of education Changjiang Scholars and innovative research team development plan "(funded by IRT1170).

【学位授予单位】：西安电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN912.35

【相似文献】