I am working on spoken digit recognition project.
Model Definition :
input_layer = Input(shape = (max_length, 1)) //Takes input vectors
input_mask = Input(shape = (max_length)) //Takes boolean mask vectors (False for zero else true)
LSTM_layer = LSTM(25)(input_layer, mask = input_mask)
dense = Dense(50, 'relu')(LSTM_layer)
dense_1 = Dense(10, 'softmax')(dense)
model = Model(inputs = [input_layer, input_mask], outputs = dense_1)
I figured it out. Just specify the dtype in both input layers as below :
input_layer = Input(shape = (max_length, 1), dtype='float32')
input_mask = Input(shape = (max_length), dtype='bool')
LSTM_layer = LSTM(25)(input_layer, mask = input_mask)
dense = Dense(50, 'relu')(LSTM_layer)
dense_1 = Dense(10, 'softmax')(dense)
model = Model(inputs = [input_layer, input_mask], outputs = dense_1)