This paper presents a text design method for Text-To-Speech synthesis application. The aim of this method is to build a corpus whose unit distribution is close to a target distribution. As text selection is a NP-hard set covering problem, a greedy algorithm is used. We propose the Kullback-Leibler divergence to compute the score of each candidate sentence. The proposed criterion gives the posibility to control the unit distribution at each step of the algorithm. Finally, we present the first results and we compare the proposed criterion with two standard criteria.