A Multidimensional Environment for the Exploration of Musical Content

In this article a multidimensional environment is defined to allow the exploration of musical content in a novel way by means of three-dimensional interfaces. The environment is created so that musical content can be located in a comprehensive space in which the world coordinates are related to music genres. To this end, the songs in a database are analyzed, and a description of their spectral content is obtained. These descriptions are then projected onto six vectors, previously determined, that represent six main genres defining a global space with six dimensions. These projected representations are useful to create a multidimensional world in which the relations, orientation, and motion will be readily intelligible to users. Music information retrieval (MIR) and music similarity evaluation have been deeply studied and discussed by the music analysis community in recent years (Downie 2006). In this context, a major portion of MIR research deals with audio recordings (as contrasted with retrieval from symbolic representations such as music notation or MIDI). The use of standard descriptors for audio-content representation has been employed widely in several works, and the descriptors have become important for music retrieval and browsing (Goto 2007). These techniques, based on the utilization of computed audio descriptors, do not suffer from the “new item” problem (i.e., a new item cannot be recommended or classified until some users rate it), and they are less likely to produce meaningless recommendations than collaborative methods (Yoshii et al. 2006). Advances in automating and simplifying the access to musical content require the analysis of the songs (Orio 2006) to obtain numerical descriptors in the time or frequency domains that can be used to measure and compare differences and similarities among the songs. (In the present article, we use the term “song” to refer to any musical piece, not only pieces sung by a vocalist. Also, when we mention “music,” we are generally referring to audio recordings of music.) In Logan (2000), a specific set of numerical descriptors—namely, mel-frequency cepstral coefficients (MFCCs)—are considered in detail Computer Music Journal, 36:3, pp. 73–83, Fall 2012 c © 2012 Massachusetts Institute of Technology. for music characterization and for the discrimination of speech and audio signals. Foote (1997) focuses on the retrieval of music and simple sound excerpts on the basis of (1) the acoustic similarity obtained using MFCCs to parametrize the audio files and (2) tree-structured quantization. In Pampalk (2006), the extraction of general music modeling parameters for music retrieval is considered. Tzanetakis and Essl (2001) and Tzanetakis and Cook (2002) focus on the classification of music samples into different meaningful genres. Aucouturier and Pachet (2002) deal with music similarity in the context of MFCCs and Gaussian characterizations. In the field of MIR, interactive music exploration is an open problem (Tzanetakis and Cook 2000; Pampalk 2001; Tzanetakis and Essl 2001) with increasing interest owing to the growing possibilities for accessing large music databases (Orio 2006). It has been observed that groups of songs show a typical clustering behavior, often related to genre taxonomy (Aucouturier and Pachet 2003). With the analysis tools at hand, music similarity has been exploited to build relations between songs or content and, ultimately, to build music-browsing interfaces in 2-D (Pampalk 2001; Pampalk and Goto 2006) and 3-D (Tzanetakis and Essl 2001; Polastre, Heyl, and Noori 2002). Other authors, like Schedl and Pohle (2010), describe approaches to explore music collections through Web pages by the use of descriptive metadata and multimedia content. In this article, a multidimensional environment for music exploration is defined on the basis of