Sonification is the representation of data in a non-verbal auditory form. It is a relatively young field, resulting from quite recent technologies, and for which there is not yet a clear standard. After a few decades of thinking about generic modeling of the concept and hundreds of case-by-case solutions, it seems interesting to make a synthesis of the commonalities and variables of all that can be called sonification in order to extract a "pattern" for this type of representation. In this thesis, we report our work on three case studies, as well as a more general study of the existing situation, aiming to lay the foundations of such a model. Our model takes the form of a function parameterized by the data and the user's interventions, whose output is the constructed sound. We propose a graphical format to represent it, before thinking about its potential as a computer design tool.