Two things can be readily said about music: musical content is multidimensional, and there is no unique way of describing musical content.

The authors of this paper describe a novel way to express musical content in a multidimensional paradigm. In the authors’ own words, this is done “so that musical content can be located in a comprehensive space in which the world coordinates are related to music genres.”

The authors analyze several songs and, using mel-frequency cepstral coefficients (MFCCs), the standard signal processing tool for music analysis, they describe the spectral characteristics of these songs. The resulting descriptors are then projected into predetermined vectors that constitute the different genres. For this purpose, using a k-means algorithm, songs of a particular genre are grouped into four clusters depending on their “signatures,” as extracted by the MFCCs. A representative signature of the genre is then calculated by finding the centroid of the most populated cluster among the four. This clustering process is repeated for each genre until six vector signatures have been developed to depict the relevant spectral behavior. In this way, a multidimensional environment is created in which “the relations, orientation, and motion [are] readily intelligible to users.”

The paper is interesting and should be well received by postgraduate students and researchers in computer music. Although I greatly appreciate the authors’ interest and ability to describe music in a multidimensional environment, it is also possible to exploit the literature on projection pursuit to view a multidimensional point in two dimensions (for example, using parallel coordinates or Andrews plots). In this process, clusters can be identified. But then, as I said in the beginning, there is no unique way of describing musical content.