A Comparative Study of Image Classification Models for Western Notation to Carnatic Notation : Conversion of Western Music Notation to Carnatic Music Notation

Abstract

Western music notation converter is essential in the field of music for the conversion of music notes which are mostly in Western staff notation to its Carnatic music note equivalent. It is difficult for the Carnatic musicians and singers to have the music notes in the form of the corresponding Swaras in Carnatic music to comprehend them. Over the past few decades, researchers have built models that recognise handwritten musical notations called Optical Music Recognition (OMR) [9]. But, when researchers who are from a non-musical background work with digital representations, the task becomes tedious and a need for processing the images arises. Therefore, instead of relying on humans for conversion of notations, image processing models are used with the help of transfer learning and classification is done using 4 models, of which 3 are pre-trained, i.e., ResNet50, VGG19, InceptionV3 and one is a simple CNN model. The models provide competitive results when compared to human experts labelling of datasets.