论文部分内容阅读
为进一步抑制噪声,提出一种基于特征值合并的语音增强算法。在经典的内嵌式预白化子空间方法的基础上,用特征值合并来提高语音质量。研究发现,对含噪语音的协方差矩阵进行特征值分解后,大特征值分量主要包含语音信息,而小特征值分量主要包含噪声,特征值分量按特征值从小到大排序后,剔除相邻的小特征值分量,可有效抑制噪声,提高语音质量。相比于其它方法,基于特征值合并的语音增强算法能有效工作于各种噪声环境中,显著提高信噪比,并有更好的语音可懂度。
In order to further suppress the noise, a speech enhancement algorithm based on eigenvalue merging is proposed. Based on the classical embedded pre-whitening subspace method, the eigenvalues are combined to improve the speech quality. After eigenvalue decomposition of the covariance matrix of noisy speech, the large eigenvalue component mainly contains speech information, while the small eigenvalue component mainly contains noise. After the eigenvalue components are sorted according to their eigenvalues, The small eigenvalue component can effectively suppress the noise and improve the voice quality. Compared with other methods, the speech enhancement algorithm based on eigenvalue merging can effectively work in various noisy environments, significantly improve the signal-noise ratio, and have better speech intelligibility.