This project was done in the winter semester of 2020 as part of the class Deep Learning for Audio Event Detection at TU Berlin.
We used https://magenta.tensorflow.org/datasets/e-gmd#download as our training and testing dataset. If you want to split the dataset into Test / Train / Val splits, use the following file: data_split.py.
After the preprocessing, the data is stored in TFRecords. In our case, we weren't able to train the models on a GPU which is why we only trained them for 20 epochs.