A semantics-driven approach to lyrics segmentation

The purpose of this paper is describing a semantics-driven approach to the automatic segmentation of song lyrics. The proposed algorithm takes into account the basic formatting commonly in use for lyrics on CD booklets and specialized Web sites, in order to extract basic semantic information, such as the organization in lines and sections. Then the algorithm applies simple rules to reconstruct lyrics structure, supporting tolerance margins as regards possible errors and encoding variants. The output is a sequence of sections labelled according to the similarity of their contents. The resulting segmenter is publicly available as a set of methods exposed via a Web application programming interface (API).