Classical music information generating system using AI Video Classification

Joo Hun Yoo, Harim Jeong, Tai-Myoung Chung

Abstract: Video classification is a machine learning method designed to analyze video format input data. Our team suggests a new online classical music watching system, providing musical information classified and generated by a depthwise convolution-based neural network model. Classical music is still limited to a few people, and there are difficult to fully enjoy the music even in the concert. Since the lack of musical knowledge is the biggest challenge, we introduce a video classification system that provides additional information to listeners in real-time. For every 50 frames of the video, our deep neural network model generated by training various classical music performance video data serves as a classifier and provides explanations of the instrument currently playing on the screen. Results of test accuracy of 95.34%, average sensitivity of 95.42%, and an average specificity of 99.68% demonstrate that efficient information can be provided to users with a low error rates. Our research team hopes that this new system, which provides additional real-time information on online classical music, will help audiences understand music deeply and even be used for educational purposes.