drumkit sound classification model used in https://codepen.io/naotokui/pen/rrGGNJ

audio_classification_spectrogram.ipynb

Thanks for sharing the code, Naotokui. :)
But I found some missing facts in your code.
In the BatchNormalization function, the axis should be -1 if the Channel is the last dimension.( if you run on Tensorflow based Keras)

And could you tell me where the augmented dataset is?
Actually the Instance size of the shared dataset(https://s3-ap-northeast-1.amazonaws.com/codepen-dev/drumkit_dataset.zip)
is different from the shown result in the notebook.
I have only 2574 number of datasets where 27069 instances are trained according to your notebook result.