Abstract
Recently, the problem of music plagiarism has emerged as an even more pressing social issue. As music information retrieval research advances, there is a growing effort to address issues related to music plagiarism. However, many studies, including our previous work, have been doing research without clearly defining what the music plagiarism detection task actually involves. This lack of a clear definition has made research progress slow and made it hard to apply results to real-world scenarios. To fix this situation, we defined how Music Plagiarism Detection is different from other MIR tasks and explained what problems need to be solved. We introduce the Similar Music Pair Dataset to support this newly defined task. In addition, we propose a method based on segment transcription as one way to solve the task.
Task Description
We define the Music Plagiarism Detection task as follows. When a full-length music A is given as a Query:
Task 1) The model should be able to select A′, which plagiarizes A from the large database.
Task 2) The model should be able to clearly specify which part of A is similar to which part of A′.
Task 3) If possible, the model should explain why the part is plagiarized (melody, vocal, or chord)
Method
Tasks 1, 2, and 3 can be accomplished in any order as long as the system is capable of performing each task effectively. Our proposed Music Plagiarism Detection system follows a comprehensive pipeline on the standard query-search based method, which integrates Music Segment Transcription with similarity analysis.
Overall architecture of our approach, which consists of three main stages: (1) Music Segment Transcription that converts raw audio into structured musical representations.
(2) Segment-level similarity analysis is employed to perform Task 2.
(3) Filtering music with segment-level similarities to perform Task 1. Task 3 can be derived depending on each similarity algorithm.
Datasets
Similar Music Pair
Sample entries from our Similar Music Pair dataset. We constructed this comprehensive music piece pair dataset for plagiarism detection evaluation. It consists of 72 pairs of original songs and comparative songs, each containing time annotations that mark where similar segments begin. Since different sound patterns can appear even within the same song pair, we constructed these parts separately and expressed them as an acoustic index. The dataset consists expressed them as an acoustic index. The dataset consists of actual plagiarism and remake cases, and contains accurate time information of similar parts for each case.
Quantitative Results
Segment-Level Performance
Segment-Level Music Plagiarism Detection results, MM stands for Multi-Modal, Music stands for music domain-based method. Best : Bold;
Music-Level Matching Segment
Our approach demonstrates a key strength in providing precise segment matching with exact temporal localization. The example of music-level detection drafts in the covers80 dataset.