Structuring music collections by exploiting peers' processing

Music collections are structured in very different ways by different useres. There is not one general taxonomy, but individual, user-specific structures exist. Most users appreciate some support in structering their collection. A large variety of methods has been developed for textual collections. However, audio data are completely different. In this paper, we present a peer to peer scenario where a music collection is enhanced a set of audio data in a node of the user's taxonomy by retrieving (partial) taxonomies of peers. In order to classify audio data into a taxonomy features need to be extracted. Adopting feature extraction to a particular set of classes is effective but not efficient. Hence, we propose again to exploit what has allready been done. Wellsuited feature extraction for one classification task is transferred to similar tasks using a new distance measures.