Speech Enhancement Using ICA with EMD-Based Reference

Different from the traditional ICA that recovers all the source signals simultaneously, the ICA with reference (ICA-R) extracts only some desired source signals from the mixtures of source signals by incorporating some a priori information into the separation process. This paper applies ICA-R to extracting a target speech signal from its noisy linear mixtures by constructing a proper reference signal with the empirical mode decomposition (EMD). Specifically, EMD is used to obtain an approximate envelope of the power spectrum of the desired speech, which is quite different from the power spectra of the environmental noises. The results of computer simulations and performance analyses demonstrate the efficiency of the proposed method.