Tacotron-2
Tacotron-2 copied to clipboard
Type of sequence_length in BiLSTM
Hi, Could you help me something about your code?
In tacotron.py in line encoder_outputs = encoder_cell(embedded_inputs, tower_input_lengths[i])
. You add tower_input_lengths[i]
(this is a tensor) to EncoderRNN (this is a BiLSTM) with code:
def __call__(self, inputs, input_lengths):
with tf.variable_scope(self.scope):
outputs, (fw_state, bw_state) = tf.nn.bidirectional_dynamic_rnn(
self._fw_cell,
self._bw_cell,
inputs,
sequence_length=input_lengths,
dtype=tf.float32,
swap_memory=True)
So sequence_length
have to int type not tensor
, but how you can do with tower_input_lengths[i]
, this is a tensor.