Skip to content
体验新版
项目
组织
正在加载...
登录
切换导航
打开侧边栏
PaddlePaddle
DeepSpeech
提交
5a632758
D
DeepSpeech
项目概览
PaddlePaddle
/
DeepSpeech
12 个月 前同步成功
通知
205
Star
8425
Fork
1598
代码
文件
提交
分支
Tags
贡献者
分支图
Diff
Issue
245
列表
看板
标记
里程碑
合并请求
3
Wiki
0
Wiki
分析
仓库
DevOps
项目成员
Pages
D
DeepSpeech
项目概览
项目概览
详情
发布
仓库
仓库
文件
提交
分支
标签
贡献者
分支图
比较
Issue
245
Issue
245
列表
看板
标记
里程碑
合并请求
3
合并请求
3
Pages
分析
分析
仓库分析
DevOps
Wiki
0
Wiki
成员
成员
收起侧边栏
关闭侧边栏
动态
分支图
创建新Issue
提交
Issue看板
体验新版 GitCode,发现更多精彩内容 >>
提交
5a632758
编写于
8月 21, 2017
作者:
X
Xinghai Sun
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
Add GRU support.
上级
638fae13
变更
7
隐藏空白更改
内联
并排
Showing
7 changed file
with
93 addition
and
12 deletion
+93
-12
demo_server.py
demo_server.py
+6
-0
evaluate.py
evaluate.py
+6
-0
infer.py
infer.py
+6
-0
layer.py
layer.py
+57
-7
model.py
model.py
+5
-4
train.py
train.py
+7
-1
tune.py
tune.py
+6
-0
未找到文件。
demo_server.py
浏览文件 @
5a632758
...
...
@@ -66,6 +66,11 @@ parser.add_argument(
default
=
512
,
type
=
int
,
help
=
"RNN layer cell number. (default: %(default)s)"
)
parser
.
add_argument
(
"--use_gru"
,
default
=
True
,
type
=
bool
,
help
=
"Use GRU or simple RNN. (default: %(default)s)"
)
parser
.
add_argument
(
"--use_gpu"
,
default
=
True
,
...
...
@@ -199,6 +204,7 @@ def start_server():
num_conv_layers
=
args
.
num_conv_layers
,
num_rnn_layers
=
args
.
num_rnn_layers
,
rnn_layer_size
=
args
.
rnn_layer_size
,
use_gru
=
args
.
use_gru
,
pretrained_model_path
=
args
.
model_filepath
)
# prepare ASR inference handler
...
...
evaluate.py
浏览文件 @
5a632758
...
...
@@ -38,6 +38,11 @@ parser.add_argument(
default
=
512
,
type
=
int
,
help
=
"RNN layer cell number. (default: %(default)s)"
)
parser
.
add_argument
(
"--use_gru"
,
default
=
True
,
type
=
bool
,
help
=
"Use GRU or simple RNN. (default: %(default)s)"
)
parser
.
add_argument
(
"--use_gpu"
,
default
=
True
,
...
...
@@ -142,6 +147,7 @@ def evaluate():
num_conv_layers
=
args
.
num_conv_layers
,
num_rnn_layers
=
args
.
num_rnn_layers
,
rnn_layer_size
=
args
.
rnn_layer_size
,
use_gru
=
args
.
use_gru
,
pretrained_model_path
=
args
.
model_filepath
)
error_rate_func
=
cer
if
args
.
error_rate_type
==
'cer'
else
wer
...
...
infer.py
浏览文件 @
5a632758
...
...
@@ -33,6 +33,11 @@ parser.add_argument(
default
=
512
,
type
=
int
,
help
=
"RNN layer cell number. (default: %(default)s)"
)
parser
.
add_argument
(
"--use_gru"
,
default
=
True
,
type
=
bool
,
help
=
"Use GRU or simple RNN. (default: %(default)s)"
)
parser
.
add_argument
(
"--use_gpu"
,
default
=
True
,
...
...
@@ -143,6 +148,7 @@ def infer():
num_conv_layers
=
args
.
num_conv_layers
,
num_rnn_layers
=
args
.
num_rnn_layers
,
rnn_layer_size
=
args
.
rnn_layer_size
,
use_gru
=
args
.
use_gru
,
pretrained_model_path
=
args
.
model_filepath
)
result_transcripts
=
ds2_model
.
infer_batch
(
infer_data
=
infer_data
,
...
...
layer.py
浏览文件 @
5a632758
...
...
@@ -57,7 +57,7 @@ def bidirectional_simple_rnn_bn_layer(name, input, size, act):
# input-hidden weights shared across bi-direcitonal rnn.
input_proj
=
paddle
.
layer
.
fc
(
input
=
input
,
size
=
size
,
act
=
paddle
.
activation
.
Linear
(),
bias_attr
=
False
)
# batch norm is only performed on input-state projection
# batch norm is only performed on input-state projection
input_proj_bn
=
paddle
.
layer
.
batch_norm
(
input
=
input_proj
,
act
=
paddle
.
activation
.
Linear
())
# forward and backward in time
...
...
@@ -68,6 +68,38 @@ def bidirectional_simple_rnn_bn_layer(name, input, size, act):
return
paddle
.
layer
.
concat
(
input
=
[
forward_simple_rnn
,
backward_simple_rnn
])
def
bidirectional_gru_bn_layer
(
name
,
input
,
size
,
act
):
"""Bidirectonal gru layer with sequence-wise batch normalization.
The batch normalization is only performed on input-state weights.
:param name: Name of the layer.
:type name: string
:param input: Input layer.
:type input: LayerOutput
:param size: Number of RNN cells.
:type size: int
:param act: Activation type.
:type act: BaseActivation
:return: Bidirectional simple rnn layer.
:rtype: LayerOutput
"""
# input-hidden weights shared across bi-direcitonal rnn.
input_proj
=
paddle
.
layer
.
fc
(
input
=
input
,
size
=
size
*
3
,
act
=
paddle
.
activation
.
Linear
(),
bias_attr
=
False
)
# batch norm is only performed on input-state projection
input_proj_bn
=
paddle
.
layer
.
batch_norm
(
input
=
input_proj
,
act
=
paddle
.
activation
.
Linear
())
# forward and backward in time
forward_gru
=
paddle
.
layer
.
grumemory
(
input
=
input_proj_bn
,
act
=
act
,
reverse
=
False
)
backward_gru
=
paddle
.
layer
.
grumemory
(
input
=
input_proj_bn
,
act
=
act
,
reverse
=
True
)
return
paddle
.
layer
.
concat
(
input
=
[
forward_gru
,
backward_gru
])
def
conv_group
(
input
,
num_stacks
):
"""Convolution group with stacked convolution layers.
...
...
@@ -83,7 +115,7 @@ def conv_group(input, num_stacks):
filter_size
=
(
11
,
41
),
num_channels_in
=
1
,
num_channels_out
=
32
,
stride
=
(
3
,
2
),
stride
=
(
2
,
2
),
padding
=
(
5
,
20
),
act
=
paddle
.
activation
.
BRelu
())
for
i
in
xrange
(
num_stacks
-
1
):
...
...
@@ -100,7 +132,7 @@ def conv_group(input, num_stacks):
return
conv
,
output_num_channels
,
output_height
def
rnn_group
(
input
,
size
,
num_stacks
):
def
rnn_group
(
input
,
size
,
num_stacks
,
use_gru
):
"""RNN group with stacked bidirectional simple RNN layers.
:param input: Input layer.
...
...
@@ -109,13 +141,25 @@ def rnn_group(input, size, num_stacks):
:type size: int
:param num_stacks: Number of stacked rnn layers.
:type num_stacks: int
:param use_gru: Use gru if set True. Use simple rnn if set False.
:type use_gru: bool
:return: Output layer of the RNN group.
:rtype: LayerOutput
"""
output
=
input
for
i
in
xrange
(
num_stacks
):
output
=
bidirectional_simple_rnn_bn_layer
(
name
=
str
(
i
),
input
=
output
,
size
=
size
,
act
=
paddle
.
activation
.
BRelu
())
if
use_gru
:
output
=
bidirectional_gru_bn_layer
(
name
=
str
(
i
),
input
=
output
,
size
=
size
,
act
=
paddle
.
activation
.
BRelu
())
else
:
output
=
bidirectional_simple_rnn_bn_layer
(
name
=
str
(
i
),
input
=
output
,
size
=
size
,
act
=
paddle
.
activation
.
BRelu
())
return
output
...
...
@@ -124,7 +168,8 @@ def deep_speech2(audio_data,
dict_size
,
num_conv_layers
=
2
,
num_rnn_layers
=
3
,
rnn_size
=
256
):
rnn_size
=
256
,
use_gru
=
True
):
"""
The whole DeepSpeech2 model structure (a simplified version).
...
...
@@ -140,6 +185,8 @@ def deep_speech2(audio_data,
:type num_rnn_layers: int
:param rnn_size: RNN layer size (number of RNN cells).
:type rnn_size: int
:param use_gru: Use gru if set True. Use simple rnn if set False.
:type use_gru: bool
:return: A tuple of an output unnormalized log probability layer (
before softmax) and a ctc cost layer.
:rtype: tuple of LayerOutput
...
...
@@ -157,7 +204,10 @@ def deep_speech2(audio_data,
block_y
=
conv_group_height
)
# rnn group
rnn_group_output
=
rnn_group
(
input
=
conv2seq
,
size
=
rnn_size
,
num_stacks
=
num_rnn_layers
)
input
=
conv2seq
,
size
=
rnn_size
,
num_stacks
=
num_rnn_layers
,
use_gru
=
use_gru
)
fc
=
paddle
.
layer
.
fc
(
input
=
rnn_group_output
,
size
=
dict_size
+
1
,
...
...
model.py
浏览文件 @
5a632758
...
...
@@ -30,9 +30,9 @@ class DeepSpeech2Model(object):
"""
def
__init__
(
self
,
vocab_size
,
num_conv_layers
,
num_rnn_layers
,
rnn_layer_size
,
pretrained_model_path
):
rnn_layer_size
,
use_gru
,
pretrained_model_path
):
self
.
_create_network
(
vocab_size
,
num_conv_layers
,
num_rnn_layers
,
rnn_layer_size
)
rnn_layer_size
,
use_gru
)
self
.
_create_parameters
(
pretrained_model_path
)
self
.
_inferer
=
None
self
.
_loss_inferer
=
None
...
...
@@ -226,7 +226,7 @@ class DeepSpeech2Model(object):
gzip
.
open
(
model_path
))
def
_create_network
(
self
,
vocab_size
,
num_conv_layers
,
num_rnn_layers
,
rnn_layer_size
):
rnn_layer_size
,
use_gru
):
"""Create data layers and model network."""
# paddle.data_type.dense_array is used for variable batch input.
# The size 161 * 161 is only an placeholder value and the real shape
...
...
@@ -243,4 +243,5 @@ class DeepSpeech2Model(object):
dict_size
=
vocab_size
,
num_conv_layers
=
num_conv_layers
,
num_rnn_layers
=
num_rnn_layers
,
rnn_size
=
rnn_layer_size
)
rnn_size
=
rnn_layer_size
,
use_gru
=
use_gru
)
train.py
浏览文件 @
5a632758
...
...
@@ -37,9 +37,14 @@ parser.add_argument(
help
=
"RNN layer number. (default: %(default)s)"
)
parser
.
add_argument
(
"--rnn_layer_size"
,
default
=
512
,
default
=
1280
,
type
=
int
,
help
=
"RNN layer cell number. (default: %(default)s)"
)
parser
.
add_argument
(
"--use_gru"
,
default
=
True
,
type
=
bool
,
help
=
"Use GRU or simple RNN. (default: %(default)s)"
)
parser
.
add_argument
(
"--adam_learning_rate"
,
default
=
5e-4
,
...
...
@@ -170,6 +175,7 @@ def train():
num_conv_layers
=
args
.
num_conv_layers
,
num_rnn_layers
=
args
.
num_rnn_layers
,
rnn_layer_size
=
args
.
rnn_layer_size
,
use_gru
=
args
.
use_gru
,
pretrained_model_path
=
args
.
init_model_path
)
ds2_model
.
train
(
train_batch_reader
=
train_batch_reader
,
...
...
tune.py
浏览文件 @
5a632758
...
...
@@ -34,6 +34,11 @@ parser.add_argument(
default
=
512
,
type
=
int
,
help
=
"RNN layer cell number. (default: %(default)s)"
)
parser
.
add_argument
(
"--use_gru"
,
default
=
True
,
type
=
bool
,
help
=
"Use GRU or simple RNN. (default: %(default)s)"
)
parser
.
add_argument
(
"--use_gpu"
,
default
=
True
,
...
...
@@ -158,6 +163,7 @@ def tune():
num_conv_layers
=
args
.
num_conv_layers
,
num_rnn_layers
=
args
.
num_rnn_layers
,
rnn_layer_size
=
args
.
rnn_layer_size
,
use_gru
=
args
.
use_gru
,
pretrained_model_path
=
args
.
model_filepath
)
# create grid for search
...
...
编辑
预览
Markdown
is supported
0%
请重试
或
添加新附件
.
添加附件
取消
You are about to add
0
people
to the discussion. Proceed with caution.
先完成此消息的编辑!
取消
想要评论请
注册
或
登录