forked from Kyubyong/tacotron
-
Notifications
You must be signed in to change notification settings - Fork 0
/
modules.py
272 lines (237 loc) · 11.2 KB
/
modules.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
# -*- coding: utf-8 -*-
#/usr/bin/python2
'''
By kyubyong park. [email protected].
https://www.github.com/kyubyong/tacotron
'''
from __future__ import print_function
from hyperparams import Hyperparams as hp
import tensorflow as tf
def embed(inputs, vocab_size, num_units, zero_pad=True, scope="embedding", reuse=None):
'''Embeds a given tensor.
Args:
inputs: A `Tensor` with type `int32` or `int64` containing the ids
to be looked up in `lookup table`.
vocab_size: An int. Vocabulary size.
num_units: An int. Number of embedding hidden units.
zero_pad: A boolean. If True, all the values of the fist row (id 0)
should be constant zeros.
scope: Optional scope for `variable_scope`.
reuse: Boolean, whether to reuse the weights of a previous layer
by the same name.
Returns:
A `Tensor` with one more rank than inputs's. The last dimesionality
should be `num_units`.
'''
with tf.variable_scope(scope, reuse=reuse):
lookup_table = tf.get_variable('lookup_table',
dtype=tf.float32,
shape=[vocab_size, num_units],
initializer=tf.truncated_normal_initializer(mean=0.0, stddev=0.01))
if zero_pad:
lookup_table = tf.concat((tf.zeros(shape=[1, num_units]),
lookup_table[1:, :]), 0)
return tf.nn.embedding_lookup(lookup_table, inputs)
def bn(inputs,
is_training=True,
activation_fn=None,
scope="bn",
reuse=None):
'''Applies batch normalization.
Args:
inputs: A tensor with 2 or more dimensions, where the first dimension has
`batch_size`. If type is `bn`, the normalization is over all but
the last dimension. Or if type is `ln`, the normalization is over
the last dimension. Note that this is different from the native
`tf.contrib.layers.batch_norm`. For this I recommend you change
a line in ``tensorflow/contrib/layers/python/layers/layer.py`
as follows.
Before: mean, variance = nn.moments(inputs, axis, keep_dims=True)
After: mean, variance = nn.moments(inputs, [-1], keep_dims=True)
is_training: Whether or not the layer is in training mode.
activation_fn: Activation function.
scope: Optional scope for `variable_scope`.
reuse: Boolean, whether to reuse the weights of a previous layer
by the same name.
Returns:
A tensor with the same shape and data dtype as `inputs`.
'''
inputs_shape = inputs.get_shape()
inputs_rank = inputs_shape.ndims
# use fused batch norm if inputs_rank in [2, 3, 4] as it is much faster.
# pay attention to the fact that fused_batch_norm requires shape to be rank 4 of NHWC.
if inputs_rank in [2, 3, 4]:
if inputs_rank == 2:
inputs = tf.expand_dims(inputs, axis=1)
inputs = tf.expand_dims(inputs, axis=2)
elif inputs_rank == 3:
inputs = tf.expand_dims(inputs, axis=1)
outputs = tf.contrib.layers.batch_norm(inputs=inputs,
center=True,
scale=True,
updates_collections=None,
is_training=is_training,
scope=scope,
fused=True,
reuse=reuse)
# restore original shape
if inputs_rank == 2:
outputs = tf.squeeze(outputs, axis=[1, 2])
elif inputs_rank == 3:
outputs = tf.squeeze(outputs, axis=1)
else: # fallback to naive batch norm
outputs = tf.contrib.layers.batch_norm(inputs=inputs,
center=True,
scale=True,
updates_collections=None,
is_training=is_training,
scope=scope,
reuse=reuse,
fused=False)
if activation_fn is not None:
outputs = activation_fn(outputs)
return outputs
def conv1d(inputs,
filters=None,
size=1,
rate=1,
padding="SAME",
use_bias=False,
activation_fn=None,
scope="conv1d",
reuse=None):
'''
Args:
inputs: A 3-D tensor with shape of [batch, time, depth].
filters: An int. Number of outputs (=activation maps)
size: An int. Filter size.
rate: An int. Dilation rate.
padding: Either `same` or `valid` or `causal` (case-insensitive).
use_bias: A boolean.
scope: Optional scope for `variable_scope`.
reuse: Boolean, whether to reuse the weights of a previous layer
by the same name.
'''
with tf.variable_scope(scope):
if padding.lower()=="causal":
# pre-padding for causality
pad_len = (size - 1) * rate # padding size
inputs = tf.pad(inputs, [[0, 0], [pad_len, 0], [0, 0]])
padding = "valid"
if filters is None:
filters = inputs.get_shape().as_list[-1]
params = {"inputs":inputs, "filters":filters, "kernel_size":size,
"dilation_rate":rate, "padding":padding, "activation":activation_fn,
"use_bias":use_bias, "reuse":reuse}
outputs = tf.layers.conv1d(**params)
return outputs
def conv1d_banks(inputs, K=16, is_training=True, scope="conv1d_banks", reuse=None):
'''Applies a series of conv1d separately.
Args:
inputs: A 3d tensor with shape of [N, T, C]
K: An int. The size of conv1d banks. That is,
The `inputs` are convolved with K filters: 1, 2, ..., K.
is_training: A boolean. This is passed to an argument of `bn`.
scope: Optional scope for `variable_scope`.
reuse: Boolean, whether to reuse the weights of a previous layer
by the same name.
Returns:
A 3d tensor with shape of [N, T, K*Hp.embed_size//2].
'''
with tf.variable_scope(scope, reuse=reuse):
outputs = conv1d(inputs, hp.embed_size//2, 1) # k=1
for k in range(2, K+1): # k = 2...K
with tf.variable_scope("num_{}".format(k)):
output = conv1d(inputs, hp.embed_size // 2, k)
outputs = tf.concat((outputs, output), -1)
outputs = bn(outputs, is_training=is_training, activation_fn=tf.nn.relu)
return outputs # (N, T, Hp.embed_size//2*K)
def gru(inputs, num_units=None, bidirection=False, scope="gru", reuse=None):
'''Applies a GRU.
Args:
inputs: A 3d tensor with shape of [N, T, C].
num_units: An int. The number of hidden units.
bidirection: A boolean. If True, bidirectional results
are concatenated.
scope: Optional scope for `variable_scope`.
reuse: Boolean, whether to reuse the weights of a previous layer
by the same name.
Returns:
If bidirection is True, a 3d tensor with shape of [N, T, 2*num_units],
otherwise [N, T, num_units].
'''
with tf.variable_scope(scope, reuse=reuse):
if num_units is None:
num_units = inputs.get_shape().as_list[-1]
cell = tf.contrib.rnn.GRUCell(num_units)
if bidirection:
cell_bw = tf.contrib.rnn.GRUCell(num_units)
outputs, _ = tf.nn.bidirectional_dynamic_rnn(cell, cell_bw, inputs, dtype=tf.float32)
return tf.concat(outputs, 2)
else:
outputs, _ = tf.nn.dynamic_rnn(cell, inputs, dtype=tf.float32)
return outputs
def attention_decoder(inputs, memory, num_units=None, scope="attention_decoder", reuse=None):
'''Applies a GRU to `inputs`, while attending `memory`.
Args:
inputs: A 3d tensor with shape of [N, T', C']. Decoder inputs.
memory: A 3d tensor with shape of [N, T, C]. Outputs of encoder network.
num_units: An int. Attention size.
scope: Optional scope for `variable_scope`.
reuse: Boolean, whether to reuse the weights of a previous layer
by the same name.
Returns:
A 3d tensor with shape of [N, T, num_units].
'''
with tf.variable_scope(scope, reuse=reuse):
if num_units is None:
num_units = inputs.get_shape().as_list[-1]
attention_mechanism = tf.contrib.seq2seq.BahdanauAttention(num_units,
memory)
decoder_cell = tf.contrib.rnn.GRUCell(num_units)
cell_with_attention = tf.contrib.seq2seq.AttentionWrapper(decoder_cell,
attention_mechanism,
num_units,
alignment_history=True)
outputs, state = tf.nn.dynamic_rnn(cell_with_attention, inputs, dtype=tf.float32) #( N, T', 16)
return outputs, state
def prenet(inputs, num_units=None, is_training=True, scope="prenet", reuse=None):
'''Prenet for Encoder and Decoder1.
Args:
inputs: A 2D or 3D tensor.
num_units: A list of two integers. or None.
is_training: A python boolean.
scope: Optional scope for `variable_scope`.
reuse: Boolean, whether to reuse the weights of a previous layer
by the same name.
Returns:
A 3D tensor of shape [N, T, num_units/2].
'''
if num_units is None:
num_units = [hp.embed_size, hp.embed_size//2]
with tf.variable_scope(scope, reuse=reuse):
outputs = tf.layers.dense(inputs, units=num_units[0], activation=tf.nn.relu, name="dense1")
outputs = tf.layers.dropout(outputs, rate=hp.dropout_rate, training=is_training, name="dropout1")
outputs = tf.layers.dense(outputs, units=num_units[1], activation=tf.nn.relu, name="dense2")
outputs = tf.layers.dropout(outputs, rate=hp.dropout_rate, training=is_training, name="dropout2")
return outputs # (N, ..., num_units[1])
def highwaynet(inputs, num_units=None, scope="highwaynet", reuse=None):
'''Highway networks, see https://arxiv.org/abs/1505.00387
Args:
inputs: A 3D tensor of shape [N, T, W].
num_units: An int or `None`. Specifies the number of units in the highway layer
or uses the input size if `None`.
scope: Optional scope for `variable_scope`.
reuse: Boolean, whether to reuse the weights of a previous layer
by the same name.
Returns:
A 3D tensor of shape [N, T, W].
'''
if not num_units:
num_units = inputs.get_shape()[-1]
with tf.variable_scope(scope, reuse=reuse):
H = tf.layers.dense(inputs, units=num_units, activation=tf.nn.relu, name="dense1")
T = tf.layers.dense(inputs, units=num_units, activation=tf.nn.sigmoid,
bias_initializer=tf.constant_initializer(-1.0), name="dense2")
outputs = H*T + inputs*(1.-T)
return outputs