基于Hilbert-Huang变换的语音增强技术研究

Study on speech enhancement based on Hilbert-Huang transform

提出了一种基于Hilbert-Huang变换的语音增强方法。首先利用经验模态分解方法(Empirical Mode Decomposition,EMD),选择合适的固有模态函数对含噪语音进行初步降噪,然后根据信噪比分别确定过减因子进行谱减运算。实验结果表明,与传统的谱减法相比,该方法的输出信噪比提高了5dB以上,尤其在非稳定噪声条件下,输出性能有更为明显的改善。经Hilbert-Huang变换后得到的特征量能较为有效地描述语音信号的非线性以及非平稳特性。

A speech enhancement method based on Hilbert-Huang transform is presented.The empirical mode decomposition method is applied to denoise initially,and then the spectral subtraction method is used to enhance each intrinsic mode function.The subtraction factor has been justified due to the variation of the SNR.The results of enhancement experiments show that,this method outperforms the standard power spectral subtraction method resulting in superior speech quality,especially in non-stationary noise environment.Hilbert-Huang method gives the true description of the non-linear and non-stationary characteristics of speech signals,and it has wide application prospect.