The Missing Link in Generative-AI Music
Major rights holders are licensing their catalogs to generative-AI music companies. But a central question remains unresolved: when a model produces a new track, which existing works influenced it? The methods intended to answer that question have not yet been validated against a shared standard, and their results remain inconsistent across systems.
Why Independent Benchmarking Matters
That is the focus of a continuing Sound Ethics university research program. In 2025, we worked with research teams at NYU Data Science, UC Riverside, and Cal Poly, each approaching music attribution from a different angle.
This year, researchers at UC Berkeley Engineering and NYU Data Science are building on that work by benchmarking attribution methods side by side under shared conditions.
Independent university labs are essential to that process, providing a neutral setting in which different systems can be tested against the same benchmarks and evaluated by the same criteria. The project asks which methods most accurately identify the works that influenced a generated track, how clearly their results can be explained, and where different approaches agree or diverge.
From Influence to Compensation
The teams are also exploring new types of generative music models that could make these influences easier to identify and report. Reliable attribution is more than a technical goal: it could provide a foundation for translating influence into compensation, helping artists and songwriters participate economically when their work shapes AI-generated music.
Related Updates
Questions?
Contact Us
© 2024 Sound Ethics Inc.
Visiting?
Finding Us
1800 Vine StreetHollywood, CA 90028, USA
Made by SF.CO
