CNN 기반 음악 장르 분류를 통한 K-POP의 J-POP과 POP 음악 융합 추세 분석

다국어 초록 (Multilingual Abstract)

This research aims to analyze the phenomenon of fusion in K-POP music with J-POP and POP genres. It was achieved by using a CNN model that classifies POP and J-POP genres and then verifying whether J-POP-like style has increased in recent K-POP music. First, the model was trained by 500 songs of each genre: J-POP and POP, which had their vocals removed. Then, six audio signal features were extracted: Spectrogram, Mel-Spectrogram, MFCC, Zero Crossing Rate, Spectral Centroid, and Chromagram. As a result, a CNN model could classify J-POP and POP music with an accuracy of approximately 90%. Afterwards, the model classified 100 K-POP songs per year, from 2020 to 2024. The results show that in 2020 POP-like music accounted for 79%, and J-POP-like music for 21%. In 2024, the proportions shifted, with songs similar to POP-like music taking up 63%, while the ones similar to J-POP-like music - 37%. In conclusion, this study suggests that in addition to the fusion of K-POP with POP music, the fusion with J-POP has increased in recent years.