def __init__(self,
out_units,
num_postnet_layers,
kernel_size,
out_channels,
is_training,
drop_rate=0.5,
trainable=True,
name=None,
dtype=None,
**kwargs):
super(PostNetV2, self).__init__(name=name,
trainable=trainable,
**kwargs)
final_conv_layer = Conv1d(kernel_size,
out_channels,
activation=None,
is_training=is_training,
drop_rate=drop_rate,
name=f"conv1d_{num_postnet_layers}",
dtype=dtype)
self.convolutions = [
Conv1d(kernel_size,
out_channels,
activation=tf.nn.tanh,
is_training=is_training,
drop_rate=drop_rate,
name=f"conv1d_{i}",
dtype=dtype) for i in range(1, num_postnet_layers)
] + [final_conv_layer]
self.projection_layer = tf.layers.Dense(out_units, dtype=dtype)
def __init__(self,
num_conv_layers,
kernel_size,
out_units,
drop_rate,
zoneout_factor_cell,
zoneout_factor_output,
is_training,
lstm_impl=LSTMImpl.LSTMCell,
trainable=True,
name=None,
dtype=None,
**kwargs):
super(EncoderV2, self).__init__(trainable=trainable,
name=name,
dtype=dtype,
**kwargs)
assert out_units % 2 == 0
self.out_units = out_units
self.zoneout_factor_cell = zoneout_factor_cell
self.zoneout_factor_output = zoneout_factor_output
self.is_training = is_training
self._lstm_impl = lstm_impl
self.convolutions = [
Conv1d(kernel_size,
out_units,
activation=tf.nn.relu,
is_training=is_training,
drop_rate=drop_rate,
name=f"conv1d_{i}",
dtype=dtype) for i in range(0, num_conv_layers)
]