In this paper we present a hybrid fuzzy-genetic algorithm for the feature and instance subset selection problem. This algorithm combines a hybrid meta-heuristic algorithm and a fuzzy self-adaptive genetic algorithm with a rotary circular crossover which is based on a half uniform crossover. The best individual in the initial population is used as initial solution of the hybrid meta-heuristic algorithm with the purpose of improving its fitness; this method is a combination of simulated annealing, taboo search and hill-climbers and allows us to speed up the convergence of the initial population. When running, the genetic algorithm adjusts its own control parameters, and the adaptability of control parameters is directed by means of two fuzzy inference systems. Besides the description of the novel evolutionary algorithm, we present the results obtained during the experiments on several known databases and on an infant cry corpus.