PDMX: A Large-Scale Public Domain MusicXML Dataset for Symbolic Music Processing
Phillip Long, Zachary Novack, Taylor Berg-Kirkpatrick, Julian McAuley
2024-09-18
Summary
This paper presents PDMX, a large-scale dataset of over 250,000 copyright-free MusicXML scores designed to support research in symbolic music processing and generative AI music systems.
What's the problem?
As generative AI music systems grow in popularity, there are increasing concerns about copyright issues and the availability of public domain music data. There is a significant shortage of freely available symbolic music data, which is essential for training and testing AI models that can understand and create music.
What's the solution?
To address this shortage, the researchers created PDMX, the largest open-source dataset of public domain MusicXML scores. They collected these scores from the score-sharing platform MuseScore and included metadata to help analyze the quality of the music. The dataset allows researchers to conduct experiments on various tasks like music generation and analysis without worrying about copyright restrictions.
Why it matters?
This research is important because it provides a valuable resource for advancing the field of music AI. By making a large, high-quality dataset available, PDMX can help researchers develop better algorithms for understanding and generating music, ultimately leading to more innovative applications in music technology.
Abstract
The recent explosion of generative AI-Music systems has raised numerous concerns over data copyright, licensing music from musicians, and the conflict between open-source AI and large prestige companies. Such issues highlight the need for publicly available, copyright-free musical data, in which there is a large shortage, particularly for symbolic music data. To alleviate this issue, we present PDMX: a large-scale open-source dataset of over 250K public domain MusicXML scores collected from the score-sharing forum MuseScore, making it the largest available copyright-free symbolic music dataset to our knowledge. PDMX additionally includes a wealth of both tag and user interaction metadata, allowing us to efficiently analyze the dataset and filter for high quality user-generated scores. Given the additional metadata afforded by our data collection process, we conduct multitrack music generation experiments evaluating how different representative subsets of PDMX lead to different behaviors in downstream models, and how user-rating statistics can be used as an effective measure of data quality. Examples can be found at https://pnlong.github.io/PDMX.demo/.