具有高可懂度的改进的维纳滤波的语音增强算法

发布时间：2018-05-25 18:16

本文选题：维纳滤波 + 先验SNR　；参考：《计算机应用与软件》2014年11期

【摘要】：提出一种具有较高可懂度的基于维纳滤波的语音增强算法。相比于其他语音增强算法,维纳滤波法可以明显提高语音质量且含有较少的音乐噪声,但是它和其他现有语音增强算法一样,都无法有效提高语音可懂度。因为维纳滤波法和其他现有算法都过多注重噪声减少,却忽略了SNR(信噪比)的估计误差和不同的语音幅度谱畸变对可懂度有更重要的影响。为改进这些缺点,此研究依据于先验SNR和增益函数来判定SNR估计误差和语音畸变区域,然后对先验SNR小于-10 d B区域的增益函数进行修正,以及幅度谱畸变大于6.02 d B区域语音进行限制。实验证明,该算法能有效提升增强后语音可懂度NCM(归一化协方差方法)的评测值。
[Abstract]:A speech enhancement algorithm based on Wiener filter with high intelligibility is proposed. Compared with other speech enhancement algorithms, Wiener filter can significantly improve speech quality and contain less music noise, but it can not effectively improve speech intelligibility as other existing speech enhancement algorithms. Because Wiener filter and other existing algorithms pay too much attention to noise reduction, but ignore SNR estimation error and different speech amplitude spectrum distortion have more important influence on intelligibility. In order to improve these shortcomings, this study is based on the prior SNR and gain function to determine the SNR estimation error and the speech distortion region, and then modifies the gain function of the priori SNR less than -10 dB region. And the amplitude spectrum distortion is larger than 6.02 dB. Experimental results show that the proposed algorithm can effectively improve the NCM (normalized covariance method) of speech intelligibility after enhancement.
【作者单位】：太原理工大学计算机科学与技术学院;
【基金】：山西省留学归国人员科研项目(2011-027) 山西省留学人员科技活动择优项目(2011-762)
【分类号】：TN912.35

【参考文献】