Generative AI Music Systems Encode Colonial Extraction: African Sonic Epistemologies Reduced to Training Data Without Consent or Reciprocity

African musical traditions operate as living knowledge systems where rhythm encodes history, spirituality, and social protocol — mbira dzavadzimu summons ancestors, talking drums transmit language, Afrobeat carries political resistance. Generative AI treats these as statistical patterns, severing sound from the communal practices, lineage obligations, and cosmological frameworks that constitute their meaning. Indigenous epistemologies (e.g., Dagara 'village as instrument,' Shona 'music as prayer') reveal that the bias is not in the output but in the ontology: AI cannot 'hear' Africa because it lacks the relational ontology that makes African music intelligible. Restoring this lens exposes 'cultural competence' as category error — competence requires participation in the knowledge system, not pattern-matching its outputs.