acegovernance and licensing of singing datasets used for AI voice synthesis and its impact on data-contributors
Source article: Revising research practices for singing data collection
Abstract As AI voice synthesis enables increasingly sophisticated vocal deepfakes and non-consensual voice cloning, the governance, licensing and access of singing datasets has become an urgent concern for data-contributors, who face significant harms from downstream and non-consensual usage of their singing data. Singing datasets are foundational to the development of high fidelity voice AI synthesis, yet current data collection practices pose challenges: data-contributors have an event-centric contribution to…
Contested: both sides are scored from claims and sources, not community votes.
"Inside the Dáil" by David Kernan, CC BY 4.0.
By July 2026, researchers examined singing data collection as AI voice synthesis advanced, analyzing three singing datasets with the Ethically Aligned Stakeholder Elicitation framework. They found data-contributors have event-centric roles with minimal authority over licensing and access, while data-collectors retain control.
The imbalance matters because singing data is increasingly treated as a freely available commodity for downstream use, exposing contributors to temporally unbounded risks of vocal deepfakes and non-consensual cloning. Uncertainty remains about what licensing steps can effectively protect personality or identity rights as synthesis capabilities continue to evolve.
- Analysis applied the Ethically Aligned Stakeholder Elicitation (EASE) framework to three singing datasets to map power-to-interest stakes.
- Data-contributors have event-centric involvement but temporally unbounded vulnerability as new AI capabilities develop.
- Authors propose revised guidelines repositioning data-contributors centrally in licensing, use and access decisions, citing personality or identity rights.
Singing datasets enable development of high-fidelity AI voice synthesis models.
Singers who contribute singing data face significant harms from downstream non-consensual use including vocal deepfakes and voice cloning.
The rundown
Researchers applied the EASE framework to three singing datasets, finding data-contributors hold minimal decision-making authority over access while data-collectors control licensing and access decisions.
The study identifies a temporal symmetry where contributor involvement is event-centric but vulnerability to harm extends beyond contribution as AI capabilities advance, prompting proposals for stakeholder authority, temporal scope, and vulnerability considerations in licensing.
Analysis is based on only three singing datasets, limiting generalizability of power-to-interest findings.
Sources
- Peer-reviewedAI & SOCIETY2026-07-12
How should this claim be treated?
The debate