A Blind Source Separation Algorithm with Linear Prediction Filters

Abstract

In this paper the theoretical analysis about why to be able to guarantee source separation by blind source separation (BSS) algorithms after linear prediction filtering is presented, and the concept of the linear prediction filter (LPF) application is illustrated. The simulation results verify the derivation of this paper and show that linear prediction analysis can be used in blind source separation for instantaneous mixtures and convolutive mixtures. For the convolutive mixtures, when the separated speech quality is improved, the separation performance is reduced a little.