Here, you will find the details for each dataset and method proposed in the paper. We recommend reading them in the following order: Drums → AudioSet → KWS, as each section builds upon the previous one. However, they can also be read in any order, though some comments may not make sense without prior context.
The plots illustrate the importance assigned by each method to different segments. Additionally, the annotated ground truth is highlighted in color, allowing for a direct comparison between the model's attributions and the actual relevant segments.
Note: The examples presented here are just a sample. If you wish to explore other cases, the code is available for execution.