In this article, we present GEREV, a guidance and assessment system for voice rehabilitation, and in particular, a system for representation and structuring of manipulated knowledge. The first point that we develop concerns details of the knowledge domain. Indeed, contrary to many Intelligent Tutoring System (ITS), systems which allow acquisition of reasoning (for example in geometry or in declarative knowledge), we work on the acquisition of the norm of acoustic parameters; these can be defined as an ensemble of physical sizes, some of wich constitute the standard intervals of production for categories such male or female. In rehabilitation, the norms must needs be adapted to student's possibilities. Therefore, we have centred a part of our work on the style and structuring of knowledge, wich will permit us to adapt the norm. Next we present our choice of representations and their implications. We qualify the quantitatives values with the help of fuzzy sets, wich allow us on the one hand to pass from the quantitative to qualitative, and the other hand to obtain as exhaustive and precise a representation as possible. At the level of the knowledge unit itself -that is the ability, defined as the entirebody of information that allows to define an acoustic parameter representation- we use frames, wich precisely describe notions of knowledge prototypesin the domain. As for the global organisation of expertise, we choose to describe it with the help of a network of frames which express relations between abilities. These combinations of representation styles are necessary in order to express the totality of knowledge details. However, they entail modifications within modules of the expert and the student which we will describe, before presenting in conclusion, an outline of thoughts on the portability of our work on other systems; and finally, perspectives which are evoked by problems concerning implantation of such a system, student assessment, and correction and guidance in GEREV.