Supplementary material for the paper
Filip Korzeniowksi and Richard Vogl, "Simple and Effective Semantic Song Segmentation", 26th International Society for Music Information Retrieval Conference (ISMIR), Daejeon, Korea, 2025.
- annotations: contains the annotations we used for training & evaluation.
- partitions: contains the partitions we used for each dataset. For training datasets, we provide cross-validation partitions (
8-fold-*.txt). For SALAMI, these partitions do not contain any tracks from other datasets such as RWC, Beatles, or Queen. We also provide the exact partitions used in Tables 3 and 4. Again, we tried to make sure to remove any overlap (see the paper for details). - predictions: predictions obtained for each model.