Assignment of Makam Labels by Hand versus Auto-Classification by Machine Learning for 325 'Hicaz Family' Pieces (FINAL REPORT OF AN INCONCLUSIVE PROJECT


Creative Commons License

Yarman O. U., Sethares W. A.

cilt.1, sa.1, ss.1-158, 2024 (Hakemsiz Dergi)

  • Yayın Türü: Makale / Vaka Takdimi
  • Cilt numarası: 1 Sayı: 1
  • Basım Tarihi: 2024
  • Sayfa Sayıları: ss.1-158
  • İstanbul Üniversitesi Adresli: Evet

Özet

An earlier study by our team involved the exploitation of "unique melodic contour features" to check whether a selection of 4 Turkish art music pieces and 4 Turkish folk music pieces from the 'Hicaz family' were indeed categorized correctly makam-wise [Yarman et al., 2020, "An Investigation of the Role of Diatonic Functions in the Seyir Phenomenon of Turkish Makam Music: Case of 'Hicaz Family'", DYUMM conference proceedings, 292-340]. We now deliver an overview of both TRT repertory labellings and tagging efforts by 6 anonymized experts versus several machine learning classifications with regards to 325 pieces pertaining to makams falling under the so-called 'Hicaz family' denomination. Our analysis is once again based on our novel "trigram melodic contour information extraction procedure", where all of the pieces operated on were transcribed according to the "SymbTR" symbolic representation database. The findings gathered with respect to this area of research seem to point to the fact that, say, Hicaz frequently blurs into kindred makams such as Uzzal and Hümayun, and the success rate in correctly determining a piece's makam never exceeds the level of about 75 percent whatever the methodology used. It is suspected that something similar must also be the case for makams of the 'Uşşak-Hüseyni family', and this field awaits further investigation. We hence suggest a new theoretical insight for identifying a piece of the repertory that is said to be in a given makam but has characteristics of other makams as well, particularly seeing as our extensive and prolonged enquiry turned out to be inconclusive and the project died out as a result.