Ensuring Traceability and Citation of AI-Generated Sound: A Minimal Metadata Framework for Cultural Heritage and Creative Industries

AbstractThe rapid diffusion of large-scale generative audio models is reshaping musical creation,distribution, and scholarship, yet it simultaneously erodes the provenance chain that underpinsintellectual credit, legal compliance, and long-term preservation. Current identification standards—ISRC, ISWC, DDEX-ERN, C2PA—were devised for human-authored or fixed-media works andprovide, at best, partial coverage of algorithmic authorship, training data lineage, or model-specificparameters. This presentation introduces MS-AIS (Minimal Set for AI-Sound), a lightweight,interoperable metadata schema designed to restore traceability and enable unambiguous citation ofAI-generated sonic artefacts across artistic research, commercial catalogues, and memoryinstitutions.Employing a combined methodology of normative gap analysis, stakeholder interviews (creators,labels, archives), and pilot implementation in Iberian sound repositories, we isolate eight mandatorydata points—persistent identifier, acoustic fingerprint, model/version, training corpus reference,prompt/seed synopsis, human operator(s), generation timestamp/location, and licence status—supplemented by optional ethical and technical descriptors.MS-AIS aligns with FAIR principles, dovetails with existing PID infrastructures (DOI, Handle),and can be serialised in JSON-LD or embedded within Broadcast Wave Format extensions,ensuring compatibility with both scholarly and industry workflows. We conclude by outlining asectoral adoption roadmap and inviting collaboration toward a formal standard within the COSTArtistic Intelligence Action. By operationalising transparency, the proposed framework safeguardscultural heritage, fosters responsible creative AI, and equips policymakers with a practical lever forevidence-based regulation.