摘要语音增强算法是一门很重要的学科，在现在语音处理方面占据了很重要的 位置，在许许多多的语音处理系统中都广泛地应用到。然而语音信号的收集与 采集只能在噪声的环境下采集。多年以来，众多研究人员致力于语音信号处理 的开发中，并且也想出了许多的方法，比如：谱减语音增强算法，维纳滤波器， 子空间算法等等。72259

在实际的处理系统操作过程中，无法避免杂音的侵入，因此我们必须对带 有噪音的语音进行分析，把多余的信号消除，恢复想要的语音。

在开发语音处理技术的时候，我们首先先对语音和噪音有所认识，这样才 能合理地处理信号，从问题的本质出发加深对语音及其噪声的研究，然后对语 音增强算法的种类进行了概括，语音质量是怎么估计的，最后本文针对其中的 谱减语音增强算法进行了深入的了解运用。

谱减语音增强算法是在人的感觉特征的基础上应用的，正因为人耳对于短 时相位不敏感，所有我们可以利用语音信号的短时幅度谱进行评估，这种算法 更加得使用于带加性噪声的语音增强。因为它的计算量很小，所以应用十分的 广泛。但是传统的谱减法它的噪声信号的频谱是通过无声段中平滑地获得的， 这样的话，估算得到的噪音功率谱与实际的噪音功率谱存在很大的差别，因此 可能会失去去噪的效果。为了能够提高去噪效果，本文就其谱减算法进行了研 究，并用了 Matlab 进行了仿真测试，验证了谱减法的有效性。

毕业论文关键词：语音增强算法 谱减算法 Matlab

Implementation of Speech Enhancement Algorithm

Abstract Speech enhancement algorithm is a very important discipline, in terms of speech processing now occupies very important position, in many speech processing systems are widely applied to。 However, speech signal collection and sampling only in noise environment。 Over the years, many researchers have devoted to the development of speech signal processing, and also come up with many methods, such as: spectrum subtraction speech enhancement algorithm, wiener filter, subspace algorithm and so on。

In the processing of actual system operation process, cannot avoid the  invasion of the noise, so we must analyze the speech with noise, the redundant signal to eliminate, restore to voice。

In the development of speech processing technology, we first to recognize speech and noise, so as to reasonably deal with signal, based on the essence of the problem to deepen the study of speech and noise, and then summarizes the kinds of speech enhancement algorithm, voice quality is how to estimate, finally based on the spectral subtraction speech enhancement algorithm is carried on the thorough understanding of the application。

Spectral subtraction speech enhancement algorithm is in the person's feeling characteristics on the basis of the application, because of the human ear is  not sensitive to short-term phase, all we can use to evaluate the amplitude spectrum of speech signal, this algorithm is more used in speech enhancement with additive noise。 Because it has a very small amount of calculation, so the application is very extensive。 But the traditional spectral subtraction of noise signal spectrum is obtained from the silent period smoothly, so, estimate the noise power spectrum and the actual noise power spectrum existence very big difference, so the denoising effect might be lost。 In order to improve the effect of denoising, in this paper, the spectral subtraction algorithm is studied, using the Matlab simulated test, verify the validity  of  the spectral subtraction。

Key words: spectrum subtraction speech enhancement algorithm Matlab algorithm