This article reports the results of two experiments in which factors such as duration, amplitude and noise are manipulated, in order to achieve more natural utterances in synthetic speech. The participants were native speakers of English, instructed to judge the naturalness of the different versions of utterances generated throughout the manipulations. The results indicate that there are significant individual preferences, as well as classification principles other than conventional ones. There is evidence to believe that further research in this area will render positive results in the search for naturalness. The same principles could be applied to search for naturalness in the prosodic structure of the synthetic utterances. Advancement in this area will surely render improvements in Spoken Dialogue Systems.