TY  - JOUR
T1  - Implementation of Music Similarity Analysis System Employing Source Separation and Automatic Transcription Learning
AU  - Gu, Yeonwoo 
AU  - Lee, Jaeho 
JO  - The Journal of Korean Institute of Communications and Information Sciences
PY  - 2023
DA  - 2023/1/14
DO  - 10.7840/kics.2023.48.11.1500
KW  - Music Similarity
KW  - Source Separation
KW  - Automatic Music Transcription
KW  - Artificial Neural Networks
AB  - Since existing music similarity analysis method are typically based on mixed music data that includes all instruments and audio channels, it is difficult to automatically detect plagiarism when there is a change in the genre or singer from the original song, such as slowing down a fast-tempo dance song and changing it to a ballad genre. In order to improve these existing similarity analysis method, this paper explored suitable open sources based on the overall structure of the music similarity analysis system proposed in previous studies, and uses them to implement an actual analysis system. To utilize open source deep learning models, we built a music dataset and preprocessed the data for training, achieving SDR of 5.439 for spleeter, a music separation model, and F1-score of 0.853 for Omnizart vocal, a music transcription model.