Within the grand goal of Electronic Music Distribution, we are interested in facilitating access to large scale databases of music titles, through content-based access methods. In this context, we will present recent research work conducted at Sony CSL in the field of high-level music descriptors. More particularly we focus on the notion of "unary descriptor", describing music titles in their entirety. We will first describe several of these descriptors and the associated automatic signal extractors designed at CSL, including rhythm and timbre. We will then sketch a novel approach in the automatic design of unary extractors, using a genetic programming approach, and will show preliminary results. We will finally illustrate the use of descriptors in the context of the Music Browser developped within the Cuidado IST project.