抄録
The purpose of this study is to propose an automatic music playlist generation system that places the listener into a positive mood. The system uses two types of playlist: mood-boosting and mood-stabilizing. Both types have four patterns, with a different structure that shifts the listener’s mood. To create a playlist, the system calculates the probability of an impression of all audio tracks by using a multinomial mixture model. In addition, the system chooses audio tracks for the playlist based on such probability. For a calculation using the model, in this study 30 types of music feature data were extracted from 1,500 pieces of sample tracks. Tagged data expressing five impressions of these tracks from 13 Japanese listeners were also received. Every four patterns of the playlist of mood-boosting and mood-stabilizing tracks were evaluated using the multiple mood scales method. For all playlists, the results showed that the indicator of a negative mood after listening to a playlist decreased from the index as compared to before the track was played. In addition, the index of liveliness in the playlists of mood-boosting tracks and the index of well-being in the playlists of mood- stabilizing tracks increased. In conclusion, the results indicate that the proposed system can design playlists of both mood-boosting and mood-stabilizing tracks that place the listener in a positive mood.
