Detecting Music Plagiarism Based on Melodic Analysis

Abstract

Copyright protection is getting more and more attention, and with an industry that is so pervasive to the general public like the music industry, the problem of plagiarism is a controversial topic. However, detecting whether a work is infringing on copyright is still a difficult problem because of the limitation in the verification work. Besides that, music plagiarism is a broad category. So there are numerous ways for an individual or organization to cause plagiarism, including sampling, rhythmic, and melodic methods. At the same time, the level for determining how many similarities between two songs is considered plagiarism is also a major obstacle. In this work, we attempted to detect music plagiarism by melodic methods: Supervised learning algorithm, Edit distance, and N-grams. We also present a new dataset of real legally-judged music plagiarism cases and conduct detailed court studies to be more objective. In progress, we conduct audits to verify which methods are really effective. Finally, we recommend the best method for developing the project’s graphical user interface. Future works will be an improvement of these methods, promising the usefulness of automatic tools that provides a measure of similarity score between songs.