Muugle: A music retrieval experimentation framework

Recently many music information retrieval (MIR) methods have been developed; most of these employ a music representation and have a notion of musical similarity. However, there is still no user-centered framework that allows the comparison of their different MIR techniques. This paper describes Muugle (Musical Utrecht University Global Lookup Engine), which is such a framework. Currently it has four interfaces for query formulation and a piano-roll editor for query modification. These interfaces and the editor are the subject of a usability study described in this paper. Feature-extraction and matching algorithms are implemented as components which can be selected by the user after query formulation. The features of the query are compared with those of the music stored in the Muugle database. As a first instantiation the database was filled with two music collections, namely 815 ringtones, and 476.621 incipits of the RISM collection. Muugle employs the vantage indexing method to enable fast retrieval results. This work concludes with the discussion of some promising higher-level feature extraction methods currently under development.