This thesis addresses the problem of creating music collectively on the basis of a shared repository of sound files. The classification of sounds in an heterogeneous database according to both a semantic taxonomy and a music oriented taxonomy based on content descriptors are analyzed, and their application to music creation is discussed. A data structure is presented that allows expressing musical ideas as combinations of sounds in the database, along with an interface where users can create and share such combinations. Finally, an interactive genetic algorithm is devised that automatically generates new combinations of sounds from existing ones influenced by ratings and preferences of listeners. The application of this concepts is demonstrated in Freesound Radio, an interactive radio station developed on top of the collaborative sound database Freesound.