Robust Preprocessing: Whitening in the Context of Blind Source Separation of Instantaneous Mixture of Audio Signals

Prewhitening is often considered a necessary but not sufficient condition for stronger stochastic independence criteria. After prewhitening the task of Blind Source Separation (BSS) become somewhat easier. Robust preprocessing involves spectral whitening, which is done by transforming correlated signals to an uncorrelated flat spectrum signal. The maximum entropy power spectrum estimation has been used as a contrast function for Independent Component Analysis (ICA). To test the separation ability for the instantaneous mixtures of speech and acoustic noise, comparative assessment with and without spectral whitening have been presented using proposed ICA algorithm and other existing algorithms. Proposed model provides SNR 42.9dB and cross correlation 0.98 for mixture of two music signals.