Speech Signals Separation Using Optimized Independent Component Analysis and Mutual Information

Abstract

One of the still problems in the Digital Signals Processing is the Blind Signal (Source) Separation (BSS). The BSS mean how to recover the original (source) signals from mixed (observed) signals via many sensors. There are many methods are used in the Blind Signal (Source) Separation problems specifically Cocktail Party problem, such as Independent Component Analysis (ICA), which has become most commonly used. Also, In more cases of the BSS problems especially the Cocktail-Party case there are number of challenges as number of mixed signals and the mixture type. In order, to enhance the performance of the ICA there are many studies for this purpose that depend on the optimization mechanisms as genetic algorithm and Particle Swarm Optimization (PSO). The advantages of a Quantum Particle Swarm Optimization (QPSO) are employed to improve the efficiency of the ICA approach using mutual information function as modern technique, which is used in de-mixing of the speech signals. In this work, a new technique is introduced, is QPSO-based ICA by using Mutual Information function as an objective function for the optimizing process. The presented method has been implemented on the real three different speech signals, with 8 KHz frequency. The results was high accuracy in the signals and more efficient in the computations requirements as the time and space which are measured by the evaluation metrics as the signal plotting, SNR, SDR, and Computation Time.