一种基于修正倒谱平滑技术改进的维纳滤波语音增强算法

传统的倒谱平滑维纳滤波算法在求取选择性平滑范围时，噪声会对维纳增益函数的倒谱产生影响，使估计出的选择性平滑范围不正确，进而导致此传统算法在平滑掉音乐噪声的同时也影响了噪声抑制效果。为此提出了一种基于修正倒谱平滑技术改进的维纳语音增强算法，该算法先用最大似然准则估计出纯净语音的倒谱，然后在纯净语音倒谱中求取基频和共振峰的位置，进而得到选择性平滑的范围。该方法提高了选择性倒谱平滑的准确性，进而改善了传统倒谱平滑维纳滤波语音增强算法的噪声抑制效果。最后在不同的噪声场景中对传统算法和改进算法进行了仿真对比，表明该算法能够在去除噪声的同时更好地保留语音的特征结构，较好地提高了带噪语音的质量。

An improved Wiener filtering speech enhancement algorithm based on modified cepstrum smooth technology

Because the noise will influence cepstrum of the gain functions when using traditional cepstrum smooth technology, it reduces the performance of noise reduction while suppressing music noise. To solve this problem, we propose a modified cepstrum smooth technology to improve Wiener filter speech enhancement algorithm. In the proposed scheme, we compute the range of selective smooth by using the estimation of clean speech cepstrum which is obtained by maximum likelihood criterion. The proposed method modifies the accuracy of selective smoothing,therefore it improves the noise reduction performance of the traditional cepstrum smooth Wiener filter speech enhancement algorithm. Finally, we simulate the proposed algorithm and compare it with traditional algorithm in different noise scenario. The simulation results show that the proposed algorithm can well reduce the noise and reserve the speech feature structure, so it has a better quality improvement of the noisy speech than traditional algorithm.