Guqing Music Recognition by Machine Learning Methods

The guqin is an ancient Chinese stringed instrument that is an important part of Chinese culture and history. Guqin notation is a unique form of tablature known as jianzi pu that is notoriously difficult to understand. Even now, several hundreds of pieces of music remain un-played because the notation is indecipherable to modern day players. In order to access this important cultural artefact and play the guqin as it was intended in ancient times, itâ–™s essential that new methods of translating and understanding the notation are developed. Professor Takashi Kuremoto leads a team at the Department of Information Technology and Media Design, Nippon Institute of Technology, Japan, that is using AI and machine learning methods to uncover the music of the past. The researchers want to utilise deep learning to automatically recognise guqin notation. This involves collaboration with academic experts from a broad range of different fields. The goal of representing guqin notation through AI and machine learning is particularly challenging because elements of music that we recognise, such as rhythm, speed and harmony are not given in jianzi pu and the title and the words of the songs needs to be considered and arranged by musicians. The researchers created a database of single jianzi pu lines which was composed of multiple handwritten images and augmented data and adopted multiple machine learning models, such as VGG16 and SVM to increase the accuracy of classification.