The Universal Music Tokenizer: A Federated Architecture and DAO Governance Framework for Culturally Sovereign Music AI

This whitepaper proposes the Universal Music Tokenizer (UMT), an open infrastructure initiative to address structural cultural bias in audio tokenization — the first and most determinative layer of any music AI pipeline. Drawing on quantitative evidence from CultureMERT (Kanatas et al., 2025), Papaioannou et al. (2025), and Nyandieka (2026), we document that dominant tokenizers systematically misrepresent non-Western musical traditions, producing Jensen-Shannon divergence scores of up to 22.6% between Western and non-Western corpora.

We present four interconnected proposals: (1) a federated codec architecture using task arithmetic to merge community-trained cultural adapters without centralising raw data; (2) a multi-college DAO governance framework with structurally differentiated voting power and community veto rights; (3) a legal architecture grounded in OCAP® principles, Community Research Agreements, and a community copyright strategy for adapter weights; and (4) a Living Music Fund channelling 15% of commercial revenue into transmission, creation, and advocacy.

Version 2.6. Working document — feedback welcome.