Keras BLSTM overfitting or wrong input shape?

Question

I created a BLSTM network for a particular purpose and i feel like something is probably done wrong. Here\'s the thing:

I have a dataset, which contains 222 audio sam