Musical culture is developed through descent-with-modification processes of complex knowledge about music creation, but how new creation styles evolve is not well understood. To study this problem, we conducted an artificial evolution experiment involving a population of machine-learning-based automatic composition models. These models evolved under evaluation by a large number of human listeners. We found that adaptive evolution of music styles can occur when the blending inheritance of high-dimensional statistics representing composition styles is incorporated into the generation update process. We also found a significant difference in musical preferences among listeners, depending on their musical experience. The results indicate the relevance of multi-parental transmission in the cultural evolution of music styles, a nontrivial fitness landscape of creation styles, and listeners’ diverse preferences.