Abstract:

Considering the existence of slight time-varying characteristics and a large number of near-zero signal values, we propose an online blind separation algorithm based on signal sample dynamic segmentation. The online algorithm of independent component analysis (ICA) is used to achieve dynamic segmentation. The FastICA algorithm is executed until converging to an optimal solution. To obtain better performance, weights of the rows in the separation matrix are adjusted. Simulation results show excellent performance of the proposed algorithm, indicating that the algorithm can effectively separate mixed acoustic signals recorded by CVR.

Key words: cockpit voice recorder (CVR),, mixed acoustic signal, on-line blind separation algorithm, dynamic segmentation