Current high-level multimedia languages are limited. Their limitation stems not from the lack of features but from the complexity caused by the excess of them and, more importantly, by their unstructured definition. Languages such as NCL, SMIL, and HTML define innumerable constructs to control the presentation of audiovisual data, but they fail to describe how these constructs relate to each other, especially in terms of behavior. There is no clear separation between basic and derived constructs, and no apparent principle of hierarchical build-up in their definition. Users may not need such principle, but it is indispensable for the people who define and implement these languages: it makes specifications and implementations manageable by reducing the language to a set of basic (primitive) concepts. In this thesis, a set of such basic concepts is proposed and taken as the language of a virtual machine for multimedia presentations. More precisely, a novel high-level multimedia language, called Smix (Synchronous Mixer), is presented and defined to serve as an appropriate abstraction layer for the definition and implementation of higher level multimedia languages. In defining Smix, that is, choosing a set of basic concepts, this work strives for minimalism but also aims at tackling major problems of current high-level multimedia languages, namely, the inadequate semantic models of their specifications and unsystematic approaches of their implementations. On the specification side, the use of a simple but expressive synchronous semantics, with a precise notion of time, is advocated. On the implementation side, a two-layered architecture that eases the mapping of specification concepts into digital signal processing primitives is proposed. The top layer (front end) is the realization of the semantics, and the bottom layer (back end) is structured as a multimedia digital signal processing dataflow.