Automated Learning of Acoustic Indexes from Data in a Text-to-Speech System for Italian