EthnoCatalogue: creating semantic descriptions of Slovene folk song and music based on melodic and metro-rhythmic analysis

 The basic research starting point of the EthnoCatalogue project is the folklore and ethnomusicological research on the verse, melodic, rhythmic, and metric features of Slovenian folk poetry. These features expose an important regularity at a high correlation between the syllabic stress structure of the verse and the rhythmic and melodic structure of the tune: a specific type of verse in a poem has a number of musical rhythmizations available, while a high percentage of the melodies of a particular rhythmic model belong to just one melody family or, in some cases, 100% to just one melody type. On the basis of a more detailed analysis of these structural elements we can, for example, establish the degree of similarity between individual melody types that can later be classified in larger melody families, or sort melodies according to different historic and style layers. These findings are presented in the GNI catalogue in the forms of melody and metric/rhythmic classification based on a special coding system of a symbolic representation of the text and melodic structure.



The EthnoCatalogue project aims to include more recent research from new technologies, especially those related to automatically obtaining information from musical sound recordings and the semantic interrelation of contents. The semantic treatment of sound recordings is problematic because their musical features are not directly evident. That is also why part of this research deals with the methods for automatically mapping the melodic structure and metric/rhythmic pattern onto the sound recording, which will enable us to automatically obtain semantic descriptions of sound recordings (especially melody and rhythm). The other part of this research will be dedicated to measuring the similarities of both semantic descriptions obtained this way and direct analysis of non-structuralized sound recordings. The objective of this research is to use machine-learning techniques to reveal the structure, rules, and relations between the archival records from the semantic descriptions and directly from the characteristics which derive from the sound, regardless of the metric/rhythmic analysis results and genre classifications. The comparison of the obtained structures with the established classifications can result in new findings about the contents in the digital archive. The results will enable broader research into the melodic and metric/rhythmic rules of Slovenian folk songs and folk music, and the application of music information retrieval methods to the genre and typological content classification in the EtnoMuza digital archive, which will represent a great advance in semantically interrelating the archive’s contents. Due to the large amount and variability of the material, proper material visualization methods should be developed. This research will be oriented towards developing proper visualization techniques for presentation on the basis of parameters that include geographic, historical, and content features of the material.