It seems that 2 times batch_normalization have been done in each CNN layer. Is it necessary? I really don't know what's the reason. Could you please explain it for me? Thanks a lot!

[yjiangling]

x.shape=(none, none, none)

output.shape = (1/2, 1/2, 1/2)

def cnn_cell (size, x, pool=True):
x = norm(conv2d(size)(x))
x = norm(conv2d(size)(x))
if pool:
x = maxpool(x)
return x

[AlphaAvatar]

I use BN at each CNN layer to reduce the loss of ctc, and also to speed up model convergence.