基于小波变换和独立分量分析的含噪混叠语音盲分离

Blind Separation of Noisy Speech Mixtures Based on Wavelet Transform and Independent Component Analysis

含噪混叠语音的分离是语音信号处理中的重要研究问题。该文针对语音信号的非平稳特性与不同语音源之间的相互独立性,提出用小波变换与独立分量分析相结合的方法来进行分离。首先利用小波变换分别对各含噪混叠语音进行消噪,然后用独立分量分析的方法对消噪后的混叠信号进行分离,最后进一步对分离信号作矢量归一和再消噪处理,得到各个语音源信号的最终估计。仿真结果表明这种方法取得了很好的分离效果。

A vital issue in speech processing is to extract source speeches from noisy mixtures. A method is presented based on wavelet transform and independent component analysis in this paper. Firstly, de-noise the noisy mixtures with discrete wavelet transform. Secondly, get them separated by independent component analysis. Finally, do the post-processing to the separated signals, then the estimated source speeches are got. Simulation results exhibit a high level of separating performance.