由于普通话测试录音环境粗糙，不可避免地引入各种各样的背景噪声，严重影响了普通话测试语料库（普测库）语音资料的使用。本文的主要目的是通过各种语音增强算法的比较，选择一种行之有效的算法对普测库语音进行去噪处理以建立一个大型、良好的汉语语料库，供其它相关环节使用，如语音合成、语音识别、语音评价等。 语音增强的目的是从带噪语音中提取尽可能纯净的原始语音。语音增强的算法多种多样。本文中的语音增强算法均基于单麦克风，并假设加性噪声和语音统计独立或不相关。本文主要介绍了语音增强领域的一些主流算法，包括以下内容： 1) 系统地研究了基于语音短时谱估计的各种增强方法，包括谱减法、维纳滤波、短时幅度谱的最小误差估计。 2) 在分析了基于语音生成模型增强方法基础上，主要研究了基于全极点模型和卡尔曼滤波方法。 3) 系统地对小波阈值去噪方法进行了研究，根据含噪语音中噪声、清音和浊音的不同特点, 重点研究了一种改进的多尺度小波域阈值去噪方法。 通过Matlab仿真、比较，结果显示：论文中实现的各种语音增强算法都在一定程度上提高了语音质量。其中，维纳滤波的效果最优，其次是卡尔曼滤波，然后是最小均方误差语音增强，最后是小波阈值去噪。 关键字：普测库，语音增强，谱减法，Wiener滤波，MMSE，小波阈值去噪，Kalman滤波

As the recording environment of Mandarin Test is rough,the degradation by all kinds of background noise is inevitable and it seriously affects the use of speech data from Mandarin Test Corpus(MTC). The goal of this thesis is to choose an effective algorithm of speech enhancement by comparision to denoise speech of MTC with aiming to establish a large and good Chinese corpus for speech processing, ranging from speech synthesis,speech recognition to speech evaluation. The goal of speech enhancement is withdrawing as far as possible the pure primitive speech from the noise speech. The approaches of speech enhancement are numerous.In this paper,we focus on enhancement from a single microphone,and assume that the noise is additive and statistically independent of the signal. We present the principles that guide researchers working in the area.What is contained in the thesis as following: (1) A variety of methods based on short-time spectrum estimation for speech enhancement are discussed,including spectral subtraction approach,the minimum mean square error estimation of the clean signal’s magnitude and Wiener filter approach. (2) In the methods based on a speech model for speech enhancement,the approach based on speech model and filter is discussed (3) wavelet threshold denoising method is discussed . According to the differences between noise, sonants and surds in the noisy speech,modified speech denoising algorithm using threshold - based wavelet in wavelet domain method is focused on. By comparision,Matlab simulation results show that through denoising processing of all kinds of speech enhancement algorirhms mentioned in the thesis,the speech quality is improved to a certain extent. the performance of Wiener filter is best,followed by Kalman filter,then MMSE,and last one is wavelet threshold denoising method. Keywords: Mandarin Test Corpus,speech enhancement,MMSE,spectral subtraction ,Wiener filter , wavelet threshold denoising method,Kalman filter