增强噪声背景中的语言信号,是信号处理领域的重要课题之一。七十年代中期以前,主要用模拟方法进行研究,较普遍的是利用语言的窄带效应进行带通滤波以改善语言的可懂度。以后随着电子计算机的广泛使用,七十年代中期,出现了用数字信号处理增强噪声中语言信号的一系列技术,并逐渐得到了应用。本文从语声信号数字处理出发,阐明在时域、频域内处理含噪语声的基本原理,叙述了自相关法、卡曼滤波、自适应滤波、线性予测滤波法,频谱平方法、基频跟踪法、最小均方差滤波、频谱减法和二阶谱法,介绍了利用语音参数模型去噪的 LMAP 估计算法和最大似然参数估计算法。对各种方法作了相应的评价,还介绍了应用前景、 It is one of the important topics in the field of signal processing to enhance the speech signal in the background of noise. Before the mid-seventies, mainly using simulation methods to study the more common is the use of narrowband effects of speech band-pass filtering to improve language intelligibility. With the widespread use of electronic computers in the future, a series of techniques for enhancing the speech signal in noise by digital signal processing appeared in the mid seventies and have been gradually applied. Based on the digital processing of speech signal, this paper illustrates the basic principle of processing noisy speech in time domain and frequency domain, and describes the basic principles of autocorrelation, Kalman filtering, adaptive filtering, linear predictive filtering, spectral flattening, Frequency tracking method, minimum mean square error filter, spectral subtraction and second-order spectral method, the LMAP estimation algorithm and maximum likelihood parameter estimation algorithm using speech parameter model are introduced. Made a corresponding assessment of various methods, also introduced the application prospects,