def make_transcription(self):
text = dict()
for trs in utils.list_directory(self.transcription_dir, abspath=True):
spk_id = os.path.splitext(os.path.basename(trs))[0]
lines = utils.open_utf8(trs, 'r').readlines()
# add utterence id from even lines starting at line 2
ids = [
spk_id + u'_' + re.sub(ur'\s+|:|;', u'', e)
for e in lines[1::2]
]
# delete linebreaks on odd lines starting at line 3
# (this does not take into account fancy unicode
# linebreaks), see
# http://stackoverflow.com/questions/3219014
transcriptions = [
def export(self):
"""Copy model files to output_dir"""
result_directory = os.path.join(self.recipe_dir, 'exp',
self.model_type)
for path in (
# exclude files starting with numbers, as we want only
# final state
p for p in utils.list_directory(result_directory, abspath=True)
if not os.path.basename(p)[0].isdigit()):
if os.path.isdir(path): # for log subdir
shutil.copytree(
path, os.path.join(self.output_dir,
os.path.basename(path)))
else:
shutil.copy(path, self.output_dir)
super(AbstractAcousticModel, self).export()
def export(self):
"""Copy the whole <recipe-dir>/decode to <output-dir>, copy
<recipe-dir>/graph to <output-dir>/graph
"""
self.log.debug('exporting results to %s', self.output_dir)
result_directory = os.path.join(self.recipe_dir, 'decode')
for path in utils.list_directory(result_directory, abspath=True):
if os.path.isdir(path):
shutil.copytree(path, os.path.join(
self.output_dir, os.path.basename(path)))
else:
shutil.copy(path, self.output_dir)
shutil.copytree(
os.path.join(self.recipe_dir, 'graph'),
os.path.join(self.output_dir, 'graph'))
super(Decode, self).export()
def correct_transcription(self):
"""Correct problems with the GlobalPhone Vietnamese transcripts
The corrections are completely ad hoc and the result are
stored in a temporary folder.
- remove trailings spaces and all double spacings and '_' from
transcriptions on every odd line but the first
- double spacings and '_' are actually only found for speakers
200 to 208
"""
# generate temporary output folder
corrected_transcription_dir = tempfile.mkdtemp()
# get the list of transcription files
trss = utils.list_directory(self.transcription_dir, abspath=True)
for trs in trss:
# read transcript file
lines = utils.open_utf8(trs, 'r').readlines()
# correct odd lines
lines[2::2] = [line.replace(u'_', u' ').replace(
u' ', u' ').strip() +
u'\n' for line in lines[2::2]]
# # write corrected version to temp
output_file = os.path.join(
corrected_transcription_dir, os.path.basename(trs))
with utils.open_utf8(output_file, 'w') as out:
for line in lines:
out.write(line)
self.transcription_dir = corrected_transcription_dir
return True