def read_data_thread_(
set_id,
sess,
input_data,
input_length,
output_data,
output_length,
enqueue_op_,
close_op_,
mean_speaker,
var_speaker,
fst):
'''Enqueue data to queue for the target domain'''
trans = tf.gfile.FastGFile(FLAGS.dann_file).readlines()
random.shuffle(trans)
for text, set_id_trans, speaker, audio_file in csv.reader(trans):
try:
text = [VOCAB_TO_INT[c]
for c in list(text)] + [VOCAB_TO_INT['</s>']]
except KeyError:
continue
if (set_id == set_id_trans and
((not FLAGS.use_train_lm) or in_fst(fst, text))):
feat = get_features(audio_file)
feat = feat - mean_speaker[speaker]
feat = feat / np.sqrt(var_speaker[speaker])
sess.run(enqueue_op_, feed_dict={
input_data: feat,
input_length: feat.shape[0],
output_data: text,
output_length: len(text)})
sess.run(close_op_)
def read_data_thread(set_id, sess, input_data, input_length, output_data,
output_length, enqueue_op, close_op, mean_speaker,
var_speaker, fst):
"""Enqueue data to queue"""
trans = FileOpen(FLAGS.trans_file).readlines()
random.shuffle(trans)
for line in trans:
line = line.strip()
text, set_id_trans, speaker, audio_file = line.split('\\')
try:
text = [VOCAB_TO_INT[c]
for c in list(text)] + [VOCAB_TO_INT['</s>']]
except KeyError:
continue
if (set_id == set_id_trans
and ((not FLAGS.use_train_lm) or in_fst(fst, text))):
feat = get_features(audio_file)
feat = feat - mean_speaker[speaker]
feat = feat / np.sqrt(var_speaker[speaker])
sess.run(enqueue_op,
feed_dict={
input_data: feat,
input_length: feat.shape[0],
output_data: text,
output_length: len(text)
})
sess.run(close_op)
def read_data_thread(
set_id,
sess,
input_data,
input_length,
output_data,
output_length,
enqueue_op,
close_op,
mean_speaker,
var_speaker,
fst):
'''Enqueue data to queue'''
trans = tf.gfile.FastGFile(FLAGS.trans_file).readlines()
random.shuffle(trans)
for text, set_id_trans, speaker, audio_file in csv.reader(trans):
text = [VOCAB_TO_INT[c] for c in list(text)]
# A space is required after the sentence due to the way FST is set up
if (text[-1] != VOCAB_TO_INT[' ']):
text.append(VOCAB_TO_INT[' '])
text.append(VOCAB_TO_INT['</s>'])
if (set_id == set_id_trans and
((not FLAGS.use_train_lm) or in_fst(fst, text))):
feat = get_features(audio_file)
feat = feat - mean_speaker[speaker]
feat = feat / np.sqrt(var_speaker[speaker])
sess.run(enqueue_op, feed_dict={
input_data: feat,
input_length: feat.shape[0],
output_data: text,
output_length: len(text)})
sess.run(close_op)