MusicGen cross-cultural study finds Japan rejects AI taste-to-music mapping

A new study published on arXiv (2608.03433) extends sonic seasoning research into AI-generated music, testing whether AI's learned taste-sound mappings work across cultures. Researchers from Italy, Japan, and Argentina used Meta's MusicGen model, fine-tuned to generate music from four taste prompts—sweet, sour, bitter, and salty. They recruited 361 participants across Argentina, Italy, and Japan, who first compared base versus fine-tuned excerpts and then rated the fine-tuned audio on twelve taste, emotion, and thermal descriptors.

The results reveal a nuanced cross-cultural picture. Preference for the fine-tuned model was confirmed in Argentina and Italy, but not in Japan, where listeners showed no significant preference. The salty prompt produced the weakest taste-sound correspondence in all three cohorts, suggesting that saltiness is the hardest gustatory concept to encode musically. When the researchers standardized ratings within each participant, the main effect of country disappeared, meaning much of the cultural difference stems from how people use rating scales rather than genuine perceptual differences. However, interactions between prompts and descriptors remained, and an exploratory factor analysis revealed that the twelve descriptors organized into different latent dimensions in each country. This indicates a structural component to cultural divergence persists.

The authors conclude that cross-cultural variation in AI-mediated sonic seasoning operates at two levels: overall magnitude of taste attribution (largely a scale-use artifact) and the relational structure of those attributions (genuine perceptual reorganization). For developers of generative music systems, this implies evaluation methods must distinguish response-style bias from actual cross-cultural differences. The study also highlights the need for multi-national testing of AI models trained on single-country datasets, especially as text-to-music tools move into global markets.