Using Psycho-Acoustic Models and Self-Organizing Maps to Create a Hierarchical Structuring of Music by Musical Styles

With the advent of large musical archives the need to provide an<br> organization of these archives becomes eminent. While artist-based<br> organizations or title indexes may help in locating a specific piece of<br> music, a more intuitive, genre-based organization is required to allow<br> users to browse an archive and explore its contents. Yet, currently these<br> organizations following musical styles have to be designed manually.<br> <br> In this paper we propose an approach to automatically create a hierarchical<br> organization of music archives following their perceived sound similarity.<br> More specifically, characteristics of frequency spectra are extracted and<br> transformed according to psycho-acoustic models. Subsequently, the Growing<br> Hierarchical Self-Organizing Map, a popular unsupervised neural network, is<br> used to create a hierarchical organization, offering both an interface for<br> interactive exploration as well as retrieval of music according to sound<br> similarity.