用子空间提高有色噪声背景下的语音质量

Using Subspace Method to Improve the Speech Quality in a Colored Noise Background

针对在语音通信过程中,纯净语音信号经常受到周围有色噪声环境干扰导致语音质量下降的问题,提出了一种用子空间提高语音质量,改善听觉感知度,降低用户听觉疲劳感的算法。该算法通过用非平稳噪声估计方法实时跟踪噪声信号的特征值,对跟踪的噪声和信号进行对角化;为了进一步消除有色噪声,对带噪语音再次进行维纳滤波处理,最后获得更新的最优特征值估计方程。实验结果表明,在信噪比不同的工厂噪声等有色噪声背景下,与传统子空间的算法相比,经过新算法的语音信噪比得到了明显的提高,且有效地降低了音乐噪声。

Aiming at the problem of the decline of speech quality resulted from the interference of the surrounding colored noise environment, an algorithm is proposed in this paper to improve the speech quality as well as the auditory perception, and to reduce the users＇ auditory fa- tigue. The algorithm obtains updated optimal eigenvalue estimation equations through the simul- taneous diagonalization of noise eigenvalue by real-time tracking using non-stationary noise esti- mation and signal eigenvalue. In addition,the noisy speech is processed by wiener filter in order to further eliminate the colored noise. Experimental results show that, in the background of different decibel colored noise, the new algorithm has apparently improved the speech signal-to-noise ratio and reduced the musical residual noise compared with the traditional subspace algorithm.