def read_issuers(file_path, corpus):
if not os.path.isfile(file_path):
return
data = jsonfile.read_json_file(file_path)
for issuer_idx, issuer_data in data.items():
issuer_type = issuer_data.get('type', None)
issuer_info = issuer_data.get('info', {})
if issuer_type == 'speaker':
gender = issuers.Gender(
issuer_data.get('gender', 'unknown').lower())
age_group = issuers.AgeGroup(
issuer_data.get('age_group', 'unknown').lower())
native_language = issuer_data.get('native_language', None)
issuer = issuers.Speaker(issuer_idx,
gender=gender,
age_group=age_group,
native_language=native_language,
info=issuer_info)
elif issuer_type == 'artist':
name = issuer_data.get('name', None)
issuer = issuers.Artist(issuer_idx,
name=name,
info=issuer_info)
else:
issuer = issuers.Issuer(issuer_idx, info=issuer_info)
corpus.import_issuers(issuer)
def load_file(self, folder_path, idx, corpus):
"""
Load speaker, file, utterance, labels
for the file with the given id.
"""
xml_path = os.path.join(folder_path, '{}.xml'.format(idx))
wav_paths = []
for wav_suffix in WAV_FILE_SUFFIXES:
wav_path = os.path.join(folder_path,
'{}_{}.wav'.format(idx, wav_suffix))
wav_name = os.path.split(wav_path)[1]
wav_idx = os.path.splitext(wav_name)[0]
if os.path.isfile(
wav_path) and wav_idx not in self.invalid_utterance_ids:
wav_paths.append(wav_path)
if len(wav_paths) == 0:
return []
with open(xml_path, 'r', encoding='utf-8') as f:
text = f.read()
transcription = TudaReader.extract_value(text, TRANSCRIPTION_PATTERN,
'transcription', xml_path)
transcription_raw = TudaReader.extract_value(
text, RAW_TRANSCRIPTION_PATTERN, 'raw_transcription', xml_path)
gender = TudaReader.extract_value(text, GENDER_PATTERN, 'gender',
xml_path)
is_native = TudaReader.extract_value(text, NATIVE_PATTERN, 'native',
xml_path)
age_class = TudaReader.extract_value(text, AGE_PATTERN, 'age',
xml_path)
speaker_idx = TudaReader.extract_value(text, SPEAKER_IDX_PATTERN,
'speaker_idx', xml_path)
if speaker_idx not in corpus.issuers.keys():
start_age_class = int(age_class.split('-')[0])
if start_age_class < 12:
age_group = issuers.AgeGroup.CHILD
elif start_age_class < 18:
age_group = issuers.AgeGroup.YOUTH
elif start_age_class < 65:
age_group = issuers.AgeGroup.ADULT
else:
age_group = issuers.AgeGroup.SENIOR
native_lang = None
if is_native == 'Ja':
native_lang = 'deu'
issuer = issuers.Speaker(speaker_idx,
gender=issuers.Gender(gender),
age_group=age_group,
native_language=native_lang)
corpus.import_issuers(issuer)
utt_ids = []
for wav_path in wav_paths:
wav_name = os.path.split(wav_path)[1]
wav_idx = os.path.splitext(wav_name)[0]
corpus.new_file(wav_path, wav_idx)
utt = corpus.new_utterance(wav_idx, wav_idx, speaker_idx)
utt.set_label_list(
annotations.LabelList.create_single(
transcription, idx=audiomate.corpus.LL_WORD_TRANSCRIPT))
utt.set_label_list(
annotations.LabelList.create_single(
transcription_raw,
idx=audiomate.corpus.LL_WORD_TRANSCRIPT_RAW))
utt_ids.append(wav_idx)
return utt_ids
def load_file(folder_path, idx, corpus):
"""
Load speaker, file, utterance, labels for the file with the given id.
"""
xml_path = os.path.join(folder_path, '{}.xml'.format(idx))
wav_paths = glob.glob(os.path.join(folder_path,
'{}_*.wav'.format(idx)))
if len(wav_paths) == 0:
return []
xml_file = open(xml_path, 'r', encoding='utf-8')
soup = BeautifulSoup(xml_file, 'lxml')
transcription = soup.recording.cleaned_sentence.string
transcription_raw = soup.recording.sentence.string
gender = soup.recording.gender.string
is_native = soup.recording.muttersprachler.string
age_class = soup.recording.ageclass.string
speaker_idx = soup.recording.speaker_id.string
if speaker_idx not in corpus.issuers.keys():
start_age_class = int(age_class.split('-')[0])
if start_age_class < 12:
age_group = issuers.AgeGroup.CHILD
elif start_age_class < 18:
age_group = issuers.AgeGroup.YOUTH
elif start_age_class < 65:
age_group = issuers.AgeGroup.ADULT
else:
age_group = issuers.AgeGroup.SENIOR
native_lang = None
if is_native == 'Ja':
native_lang = 'deu'
issuer = issuers.Speaker(speaker_idx,
gender=issuers.Gender(gender),
age_group=age_group,
native_language=native_lang)
corpus.import_issuers(issuer)
utt_ids = []
for wav_path in wav_paths:
wav_name = os.path.split(wav_path)[1]
wav_idx = os.path.splitext(wav_name)[0]
corpus.new_file(wav_path, wav_idx)
utt = corpus.new_utterance(wav_idx, wav_idx, speaker_idx)
utt.set_label_list(
annotations.LabelList.create_single(
transcription, idx=audiomate.corpus.LL_WORD_TRANSCRIPT))
utt.set_label_list(
annotations.LabelList.create_single(
transcription_raw,
idx=audiomate.corpus.LL_WORD_TRANSCRIPT_RAW))
utt_ids.append(wav_idx)
return utt_ids