Computational Modeling of Improvisation in Turkish Folk Music
Musical improvisation is a complex phenomenon, and there have been many attempts to describe and model it. Understanding a musical style through computational methods might help us to widen our knowledge in human expectation and anticipation. Moreover, predictive or generative systems based on the model of the musical style may be built. Such systems can be used as machine performers which would be able to improvise on-the-fly in interactive performances, meta-composers that would suggest improvisational ideas to other performers or as an educational tool that can help musicians to play and improvise in this particular style.
In the field of music information retrieval (MIR), most of the research is carried in Western musics, and the vast unchartered aspects of the other world musics remain as a major challenge. In order to further advance the state-of-the-art in MIR, this occidental stance should be broken, and the unique challenges brought by world musics should be considered. Further computational research into the diverse musical genres throughout the world will deepen our knowledge of universal versus genre-specific aspects of music, and allow us to truly evaluate the generality of various modeling strategies. Moreover, the findings from various cultures might open up new paths for musical creativity, expressivity and interaction.
Within this research, a new database of uzun havas (a non-metered structured improvisation form in Turkish folk music) is built with the help of Erdal Tuğcular (Gazi University Department of Music Education), and a system, which uses Variable-Length Markov Models (VLMMs) and multiple viewpoints models (MVM) to predict the melodies in the uzun hava form, is constructed. The novelty of the research lies within the representations, which take the 17-tone scale of Turkish folk music into consideration. To the best of knowledge, the work presents the first symbolic, machine-readable database of uzun havas, first usage of VLMMs and MVMs in the analysis of traditional Turkish music, and the first application of computational modeling in Turkish folk music. So far, there have been two publications regarding the research: a conference paper in ISMIR 2011 presenting the initial results, and the masters thesis approved in August, 2011. This is a project filed under the NSF CreativeIT Grant #0855758.
The Uzun Hava Humdrum Database consists of 77 songs, encompassing 10849 notes and 8 makams. It is encoded in the Humdrum **kern format. The database is aimed to help scholars from various disciplines to focus on the analysis of the Turkish folk music rather than setting up a database, and also to diversify the statistical research in ethnomusicological, cross-cultural and/or cross-genre music research, especially in research dealing with improvisation.
The computational modeling system of uzun havas is based on the multiple viewpoints modeling (MVM) framework developed at the Georgia Tech Center for Music Technology. To evaluate the system, a subset of pieces from Uzun hava Humdrum Database is picked to train the computational model. The results obtained from predictions show that the system is highly predictive in the note progressions of the transcriptions of uzun havas. This suggests that VLMMs and MVMs may be applied to makam-based and non-metered musical forms, in addition to Western musical styles.