改进的Kullback-Leibler复非负矩阵分解语音增强算法

针对单通道非负矩阵分解语音增强算法忽略相位信息的问题，提出了一种改进的Kullback-Leibler复非负矩阵分解的语音增强算法。该算法考虑到传统非负矩阵分解算法在复频域中增强语音时目标函数的影响，构建了一种适用于复频域的Kullback-Leibler散度下的目标函数，同时采用频谱一致性约束相位谱补偿算法，使其重构出的语音数据相位谱得到进一步的调制。实验结果表明，对于不同的非平稳噪声，所提出的算法在不同信噪比下均取得了较好的语音增强效果，尤其在低信噪比条件下（0 dB以下）语音增强效果较为明显，性能评估指标的增量较高，较好地克服了由传统相位谱补偿算法造成的信源失真率较低的缺点，进一步减少失真，抑制背景噪声，实现语音增强。

Speech enhancement based on improved Kullback-Leibler complex non-negative matrix factorization

Considering the problem that the single channel non-negative factorization speech enhancement algorithm neglects phase information, a speech enhancement algorithm based on improved Kullback-Leibler complex non-negative matrix factorization is proposed in this paper. This algorithm takes into account the influence of the objective function when the traditional non-negative matrix factorization (NMF) algorithm is used to enhance the speech in the complex frequency domain, an objective function under Kullback-Leibler divergence in the complex frequency domain is constructed, and the phase spectrum of the reconstructed speech data is further corrected by the phase spectrum compensation algorithm (PSC) with spectral consistency constraints. Experimental results show that the proposed algorithm has obvious speech enhancement effect under different non-stationary environments especially in low signal-to-noise ratio (below 0 dB), and the increment of performance evaluation index is higher; moreover, it can overcome the disadvantage of low source distortion rate caused by the traditional phase spectrum compensation algorithms, further reduce speech distortion and restrain background noise to realize speech enhancement.