forked from xdcesc/my_ch_speech_recognition
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtrain_self.py
146 lines (122 loc) · 7.41 KB
/
train_self.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
import os
import tensorflow as tf
from utils import get_data, data_hparams
from keras.callbacks import ModelCheckpoint
# 0.准备训练所需数据------------------------------
data_args = data_hparams()
data_args.data_type = 'train'
#data_args.data_path = '../dataset/'
data_args.self_wav = True
data_args.thchs30 = False
data_args.aishell = False
data_args.prime = False
data_args.stcmd = False
data_args.batch_size = 4
data_args.data_length = 20 #取self_wav的20条wav训练
#data_args.data_length = None #作者做实验时写小一些看效果用的，正常使用设为None，用于截取部分的lst数据
#data_args.shuffle = True
train_data = get_data(data_args)
# 0.准备验证所需数据------------------------------
data_args = data_hparams()
data_args.data_type = 'dev'
#data_args.data_path = '../dataset/'
data_args.self_wav = True
data_args.thchs30 = False
data_args.aishell = False
data_args.prime = False
data_args.stcmd = False
data_args.batch_size = 4
#data_args.data_length = None #作者做实验时写小一些看效果用的，正常使用设为None
data_args.data_length = 5
data_args.shuffle = True
dev_data = get_data(data_args)
'''
# 1.声学模型训练-----------------------------------
from model_speech.cnn_ctc import Am, am_hparams
am_args = am_hparams()
am_args.MS_OUTPUT_SIZE = len(train_data.am_vocab)
am_args.gpu_nums = 1
am_args.lr = 0.0008
am_args.is_training = True
am = Am(am_args)
if os.path.exists('logs_am/model_self.h5'):
print('load acoustic model_self...')
am.ctc_model.load_weights('logs_am/model_self.h5')
epochs = 10
batch_num = len(train_data.wav_lst) // train_data.batch_size
# checkpoint
ckpt = "model_self_{epoch:02d}-{val_acc:.2f}.hdf5"#字符串中包含格式化选项（named formatting options），epoch_num和validation_accuracy验证准确率
#“:02d”表示右对齐长度为2
#checkpoint = ModelCheckpoint(os.path.join('./checkpoint', ckpt), monitor='val_acc',
# save_weights_only=False, verbose=1, save_best_only=True)
#若出现”./”开头的参数，会从”./”开头的参数的上一个参数开始拼接。
#monitor='val_acc'监测验证准确率
#verbose详细模式，0为不打印输出信息，1位打印输出
#save_weights_only=False，不只保存权重而是整个模型
checkpoint = ModelCheckpoint(os.path.join('.\checkpoint', ckpt), monitor='val_accuracy',
save_weights_only=False, verbose=1, save_best_only=True)
#
# for k in range(epochs):
# print('this is the', k+1, 'th epochs trainning !!!')
# batch = train_data.get_am_batch()
# dev_batch = dev_data.get_am_batch()
# am.ctc_model.fit_generator(batch, steps_per_epoch=batch_num, epochs=10, callbacks=[checkpoint],
# workers=1, use_multiprocessing=False, validation_data=dev_batch, validation_steps=200)
batch = train_data.get_am_batch()
dev_batch = dev_data.get_am_batch()
am.ctc_model.fit_generator(batch, steps_per_epoch=batch_num, epochs=30, callbacks=[checkpoint],
workers=1, use_multiprocessing=False, validation_data=dev_batch, validation_steps=5)
#（训练数据生成器实例，一个epoch有几个batch(batch_num)，epoch，回调函数，最大进程数，不用多线程，验证数据生成器，validation_steps:设置验证多少次数据后取平均值作为此epoch训练后的效果）
am.ctc_model.save_weights('logs_am/model_self.h5')#保存所有层的权重，“h5”表示HDF5格式保存
#默认情况下，会以 TensorFlow 检查点文件格式保存模型的权重。权重也可以另存为 Keras HDF5 格式
'''
batch_num = len(train_data.wav_lst) // train_data.batch_size
# 2.语言模型训练-------------------------------------------
from model_language.transformer import Lm, lm_hparams
lm_args = lm_hparams()
lm_args.num_heads = 8
lm_args.num_blocks = 6
lm_args.input_vocab_size = len(train_data.pny_vocab)
lm_args.label_vocab_size = len(train_data.han_vocab)
lm_args.max_length = 100
lm_args.hidden_units = 512
lm_args.dropout_rate = 0.2
lm_args.lr = 0.0003
lm_args.is_training = True
lm = Lm(lm_args)
epochs = 10
with lm.graph.as_default():#as_default()，将此图作为运行环境的默认图
saver =tf.train.Saver()#tf.train.Saver 构造函数会针对图中所有变量将 save和restore操作 添加到图中。
with tf.Session(graph=lm.graph) as sess:#为指定图创建会话对象sess。会话会封装TensorFlow运行时的状态，并运行TensorFlow操作。
#由于 tf.Session 拥有物理资源（例如 GPU 和网络连接），因此通常（在with代码块中）
#用作上下文管理器，并在您退出代码块时自动关闭会话。
merged = tf.summary.merge_all()#将之前创建的所有总结节点（tf.summary.scalar），合并为一个操作，方便之后运行生成汇总数据
sess.run(tf.global_variables_initializer())#初始化图中的所有变量（可训练参数）
add_num = 0
if os.path.exists('logs_lm/self/checkpoint'):
print('loading language model...')
latest = tf.train.latest_checkpoint('logs_lm/self')#查找最新保存的检查点文件的文件名，latest_checkpoint(checkpoint_dir)
add_num = int(latest.split('_')[-1])#分隔后保存为列表，取最后一个
saver.restore(sess, latest)#restore(sess,save_path)，需要启动图表的会话。要恢复的变量不必初始化，因为恢复本身就是一种初始化变量的方法。
#该save_path参数通常是先前从save()调用或调用返回的值 latest_checkpoint()
writer = tf.summary.FileWriter('logs_lm/self/tensorboard', tf.get_default_graph())#FileWriter（logdir,graph）
#所有事件都会写到logdir所指的目录下
#接收到Graph对象，则会将图与张量形状信息一起可视化
for k in range(epochs):
total_loss = 0
batch = train_data.get_lm_batch()
for i in range(batch_num):
input_batch, label_batch = next(batch)
feed = {lm.x: input_batch, lm.y: label_batch}
cost,_ = sess.run([lm.mean_loss,lm.train_op], feed_dict=feed)#run方法的feed_dict参数为占位符提供具体的值
#因为run的返回和输入有相同的布局，又feed_dict是一个指令
#而不是一个张量，不会返回一个值，所以用“_”
#op:optimizer
#mean_loss:batch_mean_loss
total_loss += cost
if (k * batch_num + i) % 10 == 0:
rs=sess.run(merged, feed_dict=feed)#运行merged操作，收集汇总数据
writer.add_summary(rs, k * batch_num + i)#训练时添加总结
print('epochs', k+1, ': average loss = ', total_loss/batch_num)
saver.save(sess, 'logs_lm/self/model_%d' % (epochs + add_num))#将图中训练后的变量保存到检查点文件中
writer.close()