Harmonic Interval Organization in Georgian Traditional Polyphony: Evidence from Computational Analysis of Archival Field Recordings (1948-1973)

Abstract

This article presents our computational approach to analyzing harmonic interval distributions in traditional Georgian vocal music using a corpus of archival field recordings from the TSC Folklore Laboratory. By employing a deep learning model for performing multiple fundamental frequency (F0) estimation on polyphonic recordings of vocal music, we extract harmonic interval distributions from 2,494 polyphonic song recordings across 18 regions of Georgia. Through dimensionality reduction and clustering techniques, we identify patterns in interval usage that reflect regional musical traditions. Through verification experiments testing the accuracy of classification of the recordings’ region of origin based on the lower-dimensional embedded representations of their harmonic interval distributions as well as by the modal region of the recording’s assigned cluster, we demonstrate the predictive power of the representations of harmonic interval distribution within polyphonic recordings as extracted through our methods, offering new insights into the harmonic organization of Georgian polyphony and showcasing the potential of computational methods in ethnomusicology.