基于极值域均值模式分解和独立分量分析的语音增强方法

Speech enhancement methods based on EMMD and ICA

提出一种基于极值域均值模式分解与独立分量分析相结合的低信噪比语音增强算法,解决更多噪声环境下低信噪比语音信号增强问题。该算法的核心思想是:利用独立分量分析的特点,分离出选取的固有模态分量的固有特性,消除信息混淆。通过最大相似度,筛选出需要处理的固有模态分量,对其进行独立分量分析,使噪声特性能够进一步集中,提高最大相似度,这样更有利于噪声的滤除。由于独立分量分析存在幅值、位置的不确定性,所以对滤波后的独立分量要进行二度重构,即独立分量分析重构和极值域均值模式分解重构,得到增强后结果。

It presents a speech signal enhancement algorithm with low SNR based on EMMD and ICA in order to enhance the speech signal with low SNR in more environments. The core of the algorithm is to separate intrinsic properties of IMFs and to eliminate information confusion by ICA. To improve maximum similarity, IMFs being screened are processed by ICA, which tends to concentrate noise properties and eliminate noise. The filteied ICA needs to be constructed twice,which are ICA reconstruction and EMMD reconstruction because of the uncertained amplitude and location of ICA. Last, the enhancement results are obtained.