Transparent Music Generation

ISWS 2024 research project
International Semantic Web Research Summer School
June 9-15, Bertinoro, Italy
Supervision: Valentina Presutti

The advancement of generative artificial intelligence (AI) has introduced significant challenges in music rights management, particularly concerning the unauthorized use of existing musical works as training data. This paper addresses the critical issue of music source detection in AI-generated compositions, aiming to safeguard the intellectual property of artists, especially independent and less-known ones. We propose a novel pipeline that utilizes semantic music representations, specifically leveraging the ChoCo knowledge graph. Our methodology employs Tonal Pitch Step Distance (TPSD) to transform chord sequences into numeric representations, enabling key-independent comparisons. Key identification and chord estimation tools further refine the accuracy of detecting similarities between AI-generated music and existing compositions. To scale our approach, we implement k-means clustering on music sources, adapting sequence lengths for compatibility. Our research contributes a framework for automated music source detection, emphasizing technical challenges and future directions such as leveraging ontologies to enhance detection accuracy and support legal frameworks like the AI Act’s transparency mandates.