Evaluation of Operetta Songs Generation System Based on Impressions of Story Scenes

This paper describes a system which composes operetta songs fitting to adjectives representing producer’s impressions of story scenes. Inputs to the system are original theme music, story texts and adjectives representing producer’s impressions of story scenes. The system composes variations on theme music and lyrics based on impressions of story scenes using Kansei information processing in order to convey producer’s impression of a story to audiences. Evolutionary computation is also applied to generations of variations and lyrics. Subjects experiments are performed to verify the usefulness of the system using The Ant and the Chrysalis in Aesop’s Fables as a story. In the experiments, two types of evaluations are considered. The one is the evaluation from the viewpoint that the system generates operetta songs fitting to story scenes appropriately or not. The other is the evaluation from the viewpoint that the system generates operetta songs giving producers and listeners the same impressions of generated operetta songs or not.