Evaluation of Machine-Made Scores for Music Games Generated by a Deep Neural Network

This study has investigated some viewpoints to assess the appropriateness of machine-made scores compared with man-made scores created for Dance Dance Revolution, one of the most popular music games. It was found that the conventional software based on a machine learning approach could generate scores that looked similar to manually created ones in terms of idiom patterns of dance motion. However, it was also found that the difficulty of the machine-made scores should be controlled more appropriately to guarantee that the players enjoy the game. Based on these evaluations, improvement of the conventional software was attempted.