有色噪声下语音增强算法及应用研究

发布时间：2018-03-08 05:36

本文选题：语音增强　切入点：有色噪声　出处：《福州大学》2014年硕士论文　论文类型：学位论文

【摘要】：语音增强是从含噪语音中提取原始语音信号的技术,在很多工程领域中有广泛应用。实际中,污染噪声通常是有色噪声。目前,有色噪声下的语音增强尚需进一步研究和完善。本文通过研究经典的语音增强算法如维纳滤波、谱减法、信号子空间和卡尔曼滤波,发现它们在白噪声情形有较好的恢复效果,但在有色噪声情形,由于其统计学上的非平稳性这些语音增强算法不能有效地去除带噪语音信号中的噪声。本文提出有效的预白化技术来提高语音增强算法的去噪性能,更能有效地处理有色噪声环境下的带噪语音。本文的主要工作有两个：第一,提出一个改进的子空间方法；第二,提出了一个改进的卡尔曼滤波算法。在改进的子空间方法中,针对有色噪声我们构造出一种白化矩阵,使得原来带噪语音中的噪声变为白噪声。这个白化矩阵是基于有色噪声的AR模型来形成的。然后我们就可以利用白噪声下的子空间算法来估计白化后的语音,再通过逆白化获得原语音信号。这种方法能有效解决噪声协方差矩阵难于估计的问题。在改进的卡尔曼滤波算法中,我们同样采用预白化法来处理有色噪声干扰的语音信号。先通过白化矩阵作用于带噪语音得到白噪声下的带噪语音,再利用标准的卡尔曼滤波算法来恢复信号。与现有的卡尔曼滤波算法相比,我们并没有对卡尔曼状态转移矩阵进行增广,减少了计算量。同时,也给出一种激励噪声方差估计的方法。总之,本文提出了两个改进算法并给出了理论分析。进一步的实验仿真证实我们算法比当前热点的语音增强算法在提高语音信号质量上更有优势。另外,我们成功地将改进算法应用在信号时间延迟估计上,相比去噪前,去噪后的时间延迟估计得到较大改善,并且比现有的去噪方法更准确。
[Abstract]:Speech enhancement is a technique to extract the original speech signal from noisy speech. It is widely used in many engineering fields. In practice, the contaminated noise is usually colored noise. Speech enhancement under colored noise still needs further research and improvement. By studying classical speech enhancement algorithms such as Wiener filter, spectral subtraction, signal subspace and Kalman filter, it is found that they have better recovery effect in white noise. However, in the case of colored noise, because of its statistical non-stationary, these speech enhancement algorithms can not effectively remove the noise in noisy speech signal. In this paper, an effective prewhitening technique is proposed to improve the performance of speech enhancement algorithm. The main work of this paper is as follows: first, an improved subspace method is proposed; second, an improved Kalman filter algorithm is proposed. We construct a whitening matrix for colored noise. This whitening matrix is based on the AR model of colored noise. Then we can use the subspace algorithm under white noise to estimate the whitened speech. Then the original speech signal is obtained by inverse whitening. This method can effectively solve the problem that the noise covariance matrix is difficult to estimate. We also use the prewhitening method to process the speech signal with colored noise interference. Firstly, the white matrix is applied to the noisy speech to obtain the noisy speech under white noise. Then we use the standard Kalman filter algorithm to recover the signal. Compared with the existing Kalman filtering algorithm, we do not extend the Kalman state transition matrix and reduce the computation. A method for estimating the variance of excitation noise is also given. In this paper, two improved algorithms are proposed and the theoretical analysis is given. Further experimental results show that our algorithm is superior to the current popular speech enhancement algorithm in improving the quality of speech signal. We successfully apply the improved algorithm to signal time delay estimation. Compared with before denoising, the time delay estimation after denoising is much improved, and it is more accurate than the existing de-noising methods.
【学位授予单位】：福州大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN912.35

【相似文献】