Database Creation and Preliminary Acoustic Analysis of Mizo Folk Songs

This paper introduces the Tezpur University-Mizo Folk Song Database (TU-MFSD) which is created for the purpose of acoustic characterization and analysis of Mizo folk songs. Preliminary acoustic analysis and classification of the collected songs is carried out. The database is collected from different sources, viz., online resources, Art & Culture Department of Mizoram (A&C), All India Radio, and field recordings. Mizo folk songs belonging to five different categories are collected. Perceptual validation of the folk song categorization is carried out by 17 listeners (8 male, 9 female) of varying age groups from 18 to 60 years. F0 contour and signal energy contour, along with spectrograms are used for analyzing the characteristics of the collected Mizo folk songs. Four classification models are used for a preliminary classification to validate the effectiveness of the features. The acoustic analysis highlights the existence of creaky phonation at the end of sustained vowels in two categories i.e., Hunting chants and War chants, where distinctive dip in the pitch and energy contours is observed and validated by spectrograms. Percussions are observed to be more frequent in the lyrical regions than in the non-lyrical regions. Voicing is observed to have varying and fluctuating harmonics overlapped across different instruments. An accuracy of 87.50% is achieved using the random forest classifier, which is at par with that of existing works. Findings from the creaky phonation analysis as well as observations made from the experiments highlight the need for further investigation into the vocalization and articulation of Mizo folk songs, and thereby pave way for further studies in the characterization of Mizo folk songs.