An Analysis of Melodic Plagiarism Recognition using Musical Similarity Algorithms

Abstract

Court cases on melody infringement are decided based on vague unspecified terms like the ‘substantial similarities’ between melodies. It is then up to experts to unbiasedly explain these concepts to an untrained judge, which has often lead to controversial verdicts. This paper presents an attempt at objectifying vague terms like ‘substantial similarity’ in cases of melodic plagiarism by testing whether existing melodic similarity algorithms can help determine plagiarism. Perceived similarity by ordinary humans (intrinsic test) and expert analyses in court (extrinsic test) are used to gather requirements that an algorithm needs to meet in order to accurately predict plagiarism. Some well-known similarity algorithms and pre-processing algorithms are presented and analyzed on whether they meet the requirements. The Edit Distance algorithm does well on these requirements and is then used to compare a song to both remixes of that song and non-remixes, finding an optimal boundary between the two that we can label as a plagiarism boundary. Lastly, we discuss the performance of the algorithm and examine the potential of this approach.