Performance Analysis of Blind Source Separation Using Canonical Correlation

Abstract

Separation of blind source signals from a mixture remains an open issue. Many algorithms have been proposed for blind source separation (BSS) in the literature, but none outperforms the other. Most of the earlier BSS methods were based on the assumption that the sources are independent and non-Gaussian. From the literature, it is observed that speech signals are modelled using Gaussian models. This work focuses on a new approach for BSS in speech processing applications by considering the second-order statistics of the speech signals based on a canonical correlation approach. The performance of the algorithm is analyzed using signal-to-interference ratio, signal-to-distortion ratio, signal-to-artifact ratio and signal-to-noise ratio. Simulation results highlight the better performance of the proposed method as compared to the state of the art approaches like principal component analysis, singular value decomposition and independent component analysis algorithms.