Each .json file can be loaded as a dictionary. Keys correspond to artist IDs. Values correspond to artist names and tracks. Each track has a file ID and the vocal segments that should be classified during singer identification. Note that each artist has one validation and one test track. They also have at least 5 train tracks.