def test(self):
index = 0
next_idx = 20
for index in range(10):
next_idx, self.audio_features, self.audio_features_len, self.sparse_labels, wav_files = utils.next_batch(
next_idx,
1,
n_input,
n_context,
self.text_labels,
self.wav_files,
self.word_num_map)
print('读入语音文件: ', wav_files[0])
print('开始识别语音数据......')
d, train_ler = self.sess.run([self.decoded[0], self.label_err], feed_dict=self.get_feed_dict(dropout=1.0))
dense_decoded = tf.sparse_tensor_to_dense(d, default_value=-1).eval(session=self.sess)
dense_labels = utils.trans_tuple_to_texts_ch(self.sparse_labels, self.words)
for orig, decoded_array in zip(dense_labels, dense_decoded):
# 转成string
decoded_str = utils.trans_array_to_text_ch(decoded_array, self.words)
print('语音原始文本: {}'.format(orig))
print('识别出来的文本: {}'.format(decoded_str))
break
self.sess.close()
def test(self):
index = 0
next_idx = 20
for index in range(20):
next_idx, self.audio_features, self.audio_features_len, self.sparse_labels, wav_files = utils.next_batch(
next_idx,
1,
n_input,
n_context,
self.text_labels,
self.wav_files,
self.word_num_map)
print('读入语音文件: ', wav_files[0])
print('开始识别语音数据......')
d, train_ler = self.sess.run([self.decoded[0], self.label_err], feed_dict=self.get_feed_dict(dropout=1.0))
dense_decoded = tf.sparse_tensor_to_dense(d, default_value=-1).eval(session=self.sess)
dense_labels = utils.trans_tuple_to_texts_ch(self.sparse_labels, self.words)
for orig, decoded_array in zip(dense_labels, dense_decoded):
# 转成string
decoded_str = utils.trans_array_to_text_ch(decoded_array, self.words)
print('语音的原始文本: {}'.format(orig))
print('识别出来的文本: {}'.format(decoded_str))
break
self.sess.close()
def train(self):
epochs = 120
# 准备运行训练步骤
section = '\n{0:=^40}\n'
print(section.format('开始训练'))
train_start = time.time()
for epoch in range(epochs): # 样本集迭代次数
epoch_start = time.time()
if epoch < self.startepo:
continue
print("第：", epoch, " 次迭代，一共要迭代 ", epochs, "次")
#######################run batch####
n_batches_epoch = int(np.ceil(len(self.text_labels) / batch_size))
print("在本次迭代中一共循环： ", n_batches_epoch, "每次取：", batch_size)
train_cost = 0
train_err = 0
next_idx = 0
for batch in range(n_batches_epoch): # 一次batch_size，取多少次
# 取数据
# temp_next_idx, temp_audio_features, temp_audio_features_len, temp_sparse_labels
next_idx, self.audio_features, self.audio_features_len, self.sparse_labels, wav_files = utils.next_batch(
next_idx,
batch_size,
n_input,
n_context,
self.text_labels,
self.wav_files,
self.word_num_map)
# 计算 avg_loss optimizer ;
batch_cost, _ = self.sess.run([self.avg_loss, self.optimizer], feed_dict=self.get_feed_dict())
train_cost += batch_cost
if (batch + 1) % 70 == 0:
rs = self.sess.run(self.merged, feed_dict=self.get_feed_dict())
self.writer.add_summary(rs, batch)
print('循环次数:', batch, '损失: ', train_cost / (batch + 1))
d, train_err = self.sess.run([self.decoded[0], self.label_err], feed_dict=self.get_feed_dict(dropout=1.0))
dense_decoded = tf.sparse_tensor_to_dense(d, default_value=-1).eval(session=self.sess)
dense_labels = utils.trans_tuple_to_texts_ch(self.sparse_labels, self.words)
print('错误率: ', train_err)
for orig, decoded_array in zip(dense_labels, dense_decoded):
# convert to strings
decoded_str = utils.trans_array_to_text_ch(decoded_array, self.words)
print('语音原始文本: {}'.format(orig))
print('识别出来的文本: {}'.format(decoded_str))
break
epoch_duration = time.time() - epoch_start
log = '迭代次数 {}/{}, 训练损失: {:.3f}, 错误率: {:.3f}, time: {:.2f} sec'
print(log.format(epoch, epochs, train_cost, train_err, epoch_duration))
self.saver.save(self.sess, self.savedir + self.conf.get("FILE_DATA").savefile, global_step=epoch)
train_duration = time.time() - train_start
print('Training complete, total duration: {:.2f} min'.format(train_duration / 60))
self.sess.close()
def test_target_wav_file(self, wav_files, txt_labels):
print('读入语音文件: ', wav_files[0])
print('开始识别语音数据......')
self.audio_features, self.audio_features_len, text_vector, text_vector_len = utils.get_audio_mfcc_features(
None, wav_files, n_input, n_context, self.word_num_map, txt_labels)
self.sparse_labels = utils.sparse_tuple_from(text_vector)
d, train_ler = self.sess.run([self.decoded[0], self.label_err],
feed_dict=self.get_feed_dict(dropout=1.0))
dense_decoded = tf.sparse_tensor_to_dense(
d, default_value=-1).eval(session=self.sess)
decoded_str = utils.trans_array_to_text_ch(dense_decoded[0],
self.words)
print('语音原始文本: {}'.format(txt_labels[0]))
print('识别出来的文本: {}'.format(decoded_str))
self.sess.close()
def test_target_wav_file(self, wav_files, txt_labels):
print('读入语音文件: ', wav_files[0])
print('开始识别语音数据......')
self.audio_features, self.audio_features_len, text_vector, text_vector_len = utils.get_audio_mfcc_features(
None,
wav_files,
n_input,
n_context,
self.word_num_map,
txt_labels)
self.sparse_labels = utils.sparse_tuple_from(text_vector)
d, train_ler = self.sess.run([self.decoded[0], self.label_err], feed_dict=self.get_feed_dict(dropout=1.0))
dense_decoded = tf.sparse_tensor_to_dense(d, default_value=-1).eval(session=self.sess)
decoded_str = utils.trans_array_to_text_ch(dense_decoded[0], self.words)
print('语音原始文本: {}'.format(txt_labels[0]))
print('识别出来的文本: {}'.format(decoded_str))
self.sess.close()
def train(self):
epochs = 200
# 准备运行训练步骤
section = '\n{0:=^40}\n'
tf.logging.info(section.format('开始训练'))
train_start = time.time()
for epoch in range(epochs): # 样本集迭代次数
epoch_start = time.time()
if epoch < self.startepo:
continue
tf.logging.info("第" + str(epoch + 1) + "次迭代,一共要迭代" + str(epochs) +
"次")
#######################run batch####
n_batches_epoch = int(np.ceil(len(self.text_labels) / batch_size))
tf.logging.info("在本次迭代中一共循环" + str(n_batches_epoch) + ",每次取" +
str(batch_size))
train_cost = 0
train_err = 0
next_idx = 0
for batch in range(n_batches_epoch): # 一次batch_size，取多少次
# 取数据
# temp_next_idx, temp_audio_features, temp_audio_features_len, temp_sparse_labels
tf.logging.info(
'%d/%d:%s', batch + 1, n_batches_epoch,
datetime.datetime.now().strftime('%Y-%m-%d %H:%M:%S'))
next_idx, self.audio_features, self.audio_features_len, self.sparse_labels, wav_files = utils.next_batch(
next_idx,
batch_size,
n_input,
n_context,
self.text_labels,
self.wav_files,
self.word_num_map,
pick_deterministically=False)
# 计算 avg_loss optimizer ;
batch_cost, _ = self.sess.run([self.avg_loss, self.optimizer],
feed_dict=self.get_feed_dict())
train_cost += batch_cost
if (batch + 1) % 100 == 0:
# rs = self.sess.run(self.merged, feed_dict=self.get_feed_dict())
# self.writer.add_summary(rs, batch)
tf.logging.info('循环次数:' + str(batch + 1) + '损失:' +
str(train_cost / (batch + 1)))
d, train_err = self.sess.run(
[self.decoded[0], self.label_err],
feed_dict=self.get_feed_dict(dropout=1.0))
dense_decoded = tf.sparse_tensor_to_dense(
d, default_value=-1).eval(session=self.sess)
dense_labels = utils.trans_tuple_to_texts_ch(
self.sparse_labels, self.words)
tf.logging.info('错误率:' + str(train_err))
for orig, decoded_array in zip(dense_labels,
dense_decoded):
# convert to strings
decoded_str = utils.trans_array_to_text_ch(
decoded_array, self.words)
tf.logging.info('语音原始文本:{}'.format(
orig.encode('utf-8')))
tf.logging.info('识别出来的文本:{}'.format(
decoded_str.encode('utf-8')))
break
epoch_duration = time.time() - epoch_start
log = '迭代次数 {}/{}, 训练损失:{:.3f}, 错误率:{:.3f}, time:{:.2f} sec'
tf.logging.info(
log.format(epoch + 1, epochs, train_cost, train_err,
epoch_duration))
self.saver.save(self.sess,
self.savedir + self.conf.get("FILE_DATA").savefile,
global_step=epoch + 1)
train_duration = time.time() - train_start
tf.logging.info('Training complete, total duration:{:.2f} min'.format(
train_duration / 60))
self.sess.close()
def train(self):
epochs = 120
# 准备运行训练步骤
section = '\n{0:=^40}\n'
print(section.format('开始训练'))
train_start = time.time()
for epoch in range(epochs): # 样本集迭代次数
epoch_start = time.time()
if epoch < self.startepo:
continue
print("第：", epoch, " 次迭代，一共要迭代 ", epochs, "次")
#######################run batch####
n_batches_epoch = int(np.ceil(len(self.text_labels) / batch_size))
print("在本次迭代中一共循环： ", n_batches_epoch, "每次取：", batch_size)
train_cost = 0
train_err = 0
next_idx = 0
for batch in range(n_batches_epoch): # 一次batch_size，取多少次
# 取数据
# temp_next_idx, temp_audio_features, temp_audio_features_len, temp_sparse_labels
next_idx, self.audio_features, self.audio_features_len, self.sparse_labels, wav_files = utils.next_batch(
next_idx,
batch_size,
n_input,
n_context,
self.text_labels,
self.wav_files,
self.word_num_map)
# 计算 avg_loss optimizer ;
batch_cost, _ = self.sess.run([self.avg_loss, self.optimizer], feed_dict=self.get_feed_dict())
train_cost += batch_cost
if (batch + 1) % 70 == 0:
rs = self.sess.run(self.merged, feed_dict=self.get_feed_dict())
self.writer.add_summary(rs, batch)
print('循环次数:', batch, '损失: ', train_cost / (batch + 1))
d, train_err = self.sess.run([self.decoded[0], self.label_err], feed_dict=self.get_feed_dict(dropout=1.0))
dense_decoded = tf.sparse_tensor_to_dense(d, default_value=-1).eval(session=self.sess)
dense_labels = utils.trans_tuple_to_texts_ch(self.sparse_labels, self.words)
print('错误率: ', train_err)
for orig, decoded_array in zip(dense_labels, dense_decoded):
# convert to strings
decoded_str = utils.trans_array_to_text_ch(decoded_array, self.words)
print('语音原始文本: {}'.format(orig))
print('识别出来的文本: {}'.format(decoded_str))
break
epoch_duration = time.time() - epoch_start
log = '迭代次数 {}/{}, 训练损失: {:.3f}, 错误率: {:.3f}, time: {:.2f} sec'
print(log.format(epoch, epochs, train_cost, train_err, epoch_duration))
self.saver.save(self.sess, self.savedir + self.conf.get("FILE_DATA").savefile, global_step=epoch)
train_duration = time.time() - train_start
print('Training complete, total duration: {:.2f} min'.format(train_duration / 60))
self.sess.close()
