A method is proposed to find the digital input signals to enter in nonlinear acoustical systems (power amplifiers, transducers, etc.) in order to obtain desired arbitrary response signals. The searched input signals are found by performing a Monte Carlo search guided by a simulated annealing process applied to a hidden model with a small number of parameters. The physical system is actually used in the optimization procedure, in a real-time manner, so that no theoretical model of the system response is required. The main aspects of the algorithm are described and illustrated with several examples.