In this paper, we propose a genetic algorithm for the separation of two mixed acoustic signals in their independent components. The genetic algorithm is implemented in two variants that use different evaluation functions. The first variant uses statistical properties provided by a normalized version of the fourth moment of the probability density function of the signal, called kurtosis, while the second variant uses an evaluation based on geometric topology of the plotted points of the mixed signals. These two algorithms have been compared with two other classic ICA algorithms. The experiments were carried out using a variety of audio signals, musical sounds and male or female voices, but any other acoustic signals may be used, such as mechanical vibration of motor vehicles or other noises in the industrial environment.