Blind audio source separation using weight initialized independent component analysis

Blind audio source separation is a promising area for various applications like humanoids, human machine inter- action or adverse control mechanism, etc. ICA is a predominant approach for source separation in blind scenario. There are various versions of ICA to solve this purpose like Fast ICA, JADE, and C-ICA. The convergence speed and quality of separation is an issue. In this work a weight initialization approach is proposed for optimizing the convergence speed and experimental results reflects up to 28.57% that the proposed weight initialized ICA gives better convergence speed in comparison of Fast ICA. Here a critically determined ideal mixing system is considered where no noise component is taken into account.