-
Notifications
You must be signed in to change notification settings - Fork 195
/
transformer_wakeup_average_pooling.py
164 lines (132 loc) · 6.12 KB
/
transformer_wakeup_average_pooling.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
# coding=utf-8
# Copyright (C) 2022 ATHENA AUTHORS; Yanguang Xu; Yang Han; Jianwei Sun
# All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
# Only support eager mode
# pylint: disable=no-member, invalid-name, relative-beyond-top-level
# pylint: disable=too-many-locals, too-many-statements, too-many-arguments, too-many-instance-attributes
""" speech transformer implementation"""
from absl import logging
import tensorflow as tf
from .base import BaseModel
from ...utils.hparam import register_and_parse_hparams
from ...layers.transformer import TransformerEncoderLayer, TransformerEncoder
from ...layers.commons import PositionalEncoding
class KWSTransformer_AveragePooling(BaseModel):
""" Standard implementation of a KWSTransformer. Model mainly consists of three parts:
the x_net for input preparation, the transformer itself
"""
default_config = {
"num_classes":1,
"num_filters": 512,
"d_model": 512,
"num_heads": 8,
"num_encoder_layers": 12,
"rate": 0.1
}
def __init__(self, data_descriptions, config=None):
super().__init__()
self.hparams = register_and_parse_hparams(self.default_config, config, cls=self.__class__)
layers = tf.keras.layers
self.loss_function = tf.keras.losses.BinaryCrossentropy(from_logits=True)
self.metric = tf.keras.metrics.BinaryAccuracy()
# learnable class embedding to learn a global feature that represents the whole spectrogram
self.class_emb = self.add_weight("class_emb", shape=(1, 1, self.hparams.d_model), initializer=tf.keras.initializers.TruncatedNormal(mean=0., stddev=0.02))
input_features = layers.Input(
shape=data_descriptions.sample_shape["input"],
dtype=tf.float32)
# tf.ensure_shape(input_features, [None, None, 63, 1])
inner = layers.Conv2D(
filters=self.hparams.num_filters,
kernel_size=(3, 3),
strides=(2, 2),
padding="same",
use_bias=False,
data_format="channels_last",
)(input_features)
# tf.ensure_shape(inner, [None, None, 32, 512])
inner = layers.BatchNormalization()(inner)
inner = tf.nn.relu6(inner)
inner = layers.Conv2D(
filters=self.hparams.num_filters,
kernel_size=(3, 3),
strides=(2, 2),
padding="same",
use_bias=False,
data_format="channels_last",
)(inner)
inner = layers.BatchNormalization()(inner)
# tf.ensure_shape(inner, [None, None, 16, 512])
inner = tf.nn.relu6(inner)
_, _, dim, channels = inner.get_shape().as_list()
output_dim = dim * channels
inner = layers.Reshape((-1, output_dim))(inner)
inner = layers.Dense(self.hparams.d_model, activation=tf.nn.relu6)(inner)
self.x_net = tf.keras.Model(inputs=input_features, outputs=inner, name="x_net")
logging.info(self.x_net.summary())
self.concat_layer = layers.Concatenate(axis=1)
self.dropout_layer = layers.Dropout(self.hparams.rate)
# transformer encoder
encoder_layers = [
TransformerEncoderLayer(self.hparams.d_model, self.hparams.num_heads) for _ in range(self.hparams.num_encoder_layers)
]
self.encoder = TransformerEncoder(encoder_layers)
self.dense_layer = layers.Dense(self.hparams.d_model, input_shape=(self.hparams.d_model,))
# last layer for output
self.final_layer = layers.Dense(self.hparams.num_classes, input_shape=(self.hparams.d_model,))
self.tflite_model = self.build_model(data_descriptions)
def call(self, samples, training=None):
x0 = samples["input"]
x = self.x_net(x0, training=training)
batch_size = tf.shape(x)[0]
class_emb = tf.broadcast_to(self.class_emb, [batch_size, 1, self.hparams.d_model])
x = self.concat_layer([class_emb, x])
x = PositionalEncoding(self.hparams.d_model, scale=False)(x)
x = self.dropout_layer(x,training=training) # self.hparams.rate
x = self.encoder(x, training=training)
#pool_size = tf.shape(x)[1].numpy()
x_ = tf.keras.layers.AveragePooling1D(
pool_size=None,
strides=1,
padding='valid',
data_format='channels_last',
)(x)
y_ = self.dense_layer(x_)
y = self.final_layer(y_)
return y
def build_model(self, data_descriptions):
input_features = tf.keras.layers.Input(
shape= data_descriptions.sample_shape["input"],
dtype=tf.float32,
name="input_features")
x = self.x_net(input_features, training=False)
batch_size = tf.shape(x)[0]
class_emb = tf.broadcast_to(self.class_emb, [batch_size, 1, self.hparams.d_model])
x = self.concat_layer([class_emb, x])
x = PositionalEncoding(self.hparams.d_model, scale=False)(x)
x = self.dropout_layer(x,training=False) # self.hparams.rate
x = self.encoder(x, training=False)
#pool_size = tf.shape(x)[1].numpy()
x_ = tf.keras.layers.AveragePooling1D(
pool_size=None,
strides=1,
padding='valid',
data_format='channels_last',
)(x)
y_ = self.dense_layer(x_)
logits = self.final_layer(y_)
prob = tf.math.sigmoid(logits, name="sigmoid")
tflite_model = tf.keras.Model(inputs=input_features, outputs=prob, name="kws")
return tflite_model