【作者】 谭乔来；

【导师】 钱盛友；

【作者基本信息】 湖南师范大学 ， 电路与系统， 2008， 硕士

【摘要】 语音增强的目的是从带噪语音信号中提取尽可能纯净的原始语音信号,在语音识别与编码、语音通信等领域中有着广泛的应用,现已成为语音信号处理的一个重要分支。本文对受加性噪声污染的语音信号增强方法进行了研究,主要工作和创新点如下:(1)对谱相减法、维纳滤波法、卡尔曼滤波法等语音增强方法的基本原理进行了分析,并比较了这些方法的优缺点。(2)论述了小波变换的基本理论,介绍了常用的小波函数,分析了语音信号和噪声信号在小波分解后的不同特性,阐述了小波去噪的原理和方法,给出了一种改进阈值函数的小波去噪方法。实验仿真表明该方法与传统小波去噪方法相比,能获得较满意的去噪效果。(3)论述了小波包变换以及小波包去噪的原理,通过实验与小波去噪比较验证了小波包去噪的有效性。由于用小波包对染噪语音信号进行分析时,能进一步细分信号的高频部分,对高频部分的信号进行更精细的处理,从而能提高语音去噪后的信噪比。论述了多小波变换的原理以及多小波的常用预处理方法,将常用于图像处理的多小波去噪方法引入到语音信号去噪中,实验证实了多小波语音去噪的有效性。(4)论述了信号子空间法语音增强原理,并将其应用到端点检测中,给出了一种基于信号子空间和信息复杂度相结合的语音端点检测方法,给出了一种新的判断语音与非语音的信息复杂度的门限。实验仿真表明,该方法相对传统的语音端点检测方法,能提高语音端点检测准确率,特别在低信噪比条件下具有较高的端点检测准确率。更多还原

【Abstract】 The purpose of speech enhancement is extracting speeeh information from noisy speeeh signal with might and main, which is extensively used in speeeh reeognition, speeeh coding, speeeh communication and other fields.speech enhancement is becoming an important branch of speech signal processing now.This paper mainly studied the enhancement methods of speeeh signal which is polluted by additive noise . The main works and innovations are as follows:(1) The basic principles of several speech enhancement methods are discussed such as spectral subtraction method, Wiener filtering method,Kalman filtering method.The strongpoint and shortcoming of the above methods are also compared with each other in this paper.(2)The basic principle of wavelet transform is discussed,and the common wavelet function is introduced.The different characters of wavelet analysis coefficients between speech signal and noise signal are analysed,and the principle of wavelet denoising is expatiated.A new method with an improved threshold function is proposed in this paper. Experimental results show that a satisfying de-noising result can be obtained by this method compared with traditional methods.(3)The thoeries of wavelet packet transform and wavelet packet denoising are discussed,the experimnetal sumulations show the validity of this method compared with wavelet denoising,because wavelet packet ananlysis can decompose and process high-frequence signal,and a higher output SNR can be obtained.The thoeries of multi-wavelet transform and the common preprocessing methods are also introduced,which are usually used to denoise in image processing.We used the multi-wavelet thoery in speech denoising, the validity of this method can be proved by experimental sumulations.(4)The principles of speech enhancement by signal subspace are discussed,which are used in speech endpoint detection.A new speech endpoint detection method based on signal subspace and the information complexity is proposed,and an improved threshold is given to compute the information complexity to decide which is speech or not. Experimental simulations show that this method can increase the endpoint detection accuracy ratio compared with the traditional speech endpoint detection method, especially it has high endpoint detection accuracy ratio under low signal-to-noise ratio.更多还原