We study the performance gain in Blind Source Separation and Deconvolution (BSSD) systems when Acoustic Source Localization (ASL) has been employed as a pre-processing step. The study is organized using performance measures that are functions of the accuracy of the ASL system and the reverberation level. We specifically investigate the cases (a) where the actual positions of the sources are known (b) where the ASL system generates estimates subject to errors. Simulations are performed using actual BSSD systems. Using derived performance measures we show that performance of a BSSD system that knows the position of the sources accurately, benefits from this apriori knowledge.
展开▼
