一种基于最小统计噪声估计的改进谱减法

An improved spectral subtraction method based on minimum statistics noise estimation

针对传统最小统计噪声估计谱减法在低信噪非平稳噪声环境下存在的噪声估计误差较大及搜索延迟较大的问题,对带噪语音进行频域平滑然后进行全局平滑,并且提出使用双向搜索的方法,减小了搜索延迟。实验结果表明,文中提出的方法有效地实现了语音增强,具有一定的使用价值。

The big estimation error and the big delay for noise are two important problems in using the traditional spectral subtraction based on the estimation of the statistics of spectral minima for the low signal to noise ratio （SNR） input and nonstationary noise environment. In order to improve the accuracy of the estimation of noise and reduce the delay, the first frequency domain smoothing and then overall smoothing and bidirectional search are presented in this paper. The experiments prove that the method can effectively improve the intelligibility of speech. So it is a practical method.