def initialize_new(self, scp_list, word_mlf, dict, remove_previous=False):
System.set_log_dir(self.name)
if remove_previous:
for f in glob.iglob(System.get_log_dir() + '/*'):
os.remove(f)
if not remove_previous and (
os.path.exists(self.train_files_dir) or
len(glob.glob(self.model_dir + '/' + self.name + '.*')) > 0):
raise ExistingFilesException
if os.path.exists(self.train_files_dir):
shutil.rmtree(self.train_files_dir)
for f in glob.iglob(self.model_dir + '/' + self.name + '.*'):
os.remove(f)
os.mkdir(self.train_files_dir)
# handle dictionary
dic = HTK_dictionary()
if isinstance(dict, basestring):
dic.read_dict(dict)
elif all(isinstance(d, basestring) for d in dict):
for d in dict:
dic.read_dict(d)
else:
raise TypeError
dic.write_dict(self.training_dict)
self.phones = dic.get_phones()
# handle transcription
trans = HTK_transcription()
# if isinstance(word_mlf,basestring):
# trans.read_mlf(word_mlf, HTK_transcription.WORD)
# elif all(isinstance(w,basestring) for w in word_mlf):
# for w in word_mlf:
# trans.read_mlf(w, HTK_transcription.WORD)
# else:
# raise TypeError
word_mlf = word_mlf.strip().split(',')
for w in word_mlf:
trans.read_mlf(w, HTK_transcription.WORD)
self.id = 1
phones_list = self._get_model_name_id() + '.hmmlist'
with open(phones_list, 'w') as phones_desc:
for p in self.phones:
print(p, file=phones_desc)
# handle scp files
scp_list = scp_list.strip().split(',')
# if isinstance(scp_list,basestring):
# scp_list = [scp_list]
real_trans = HTK_transcription()
real_trans.transcriptions[real_trans.WORD] = {}
with open(self.training_scp, 'w') as scp_desc:
for scp in scp_list:
for file in open(scp):
id = os.path.splitext(os.path.basename(file.strip()))[0]
if not file.startswith('/'):
file = os.path.join(os.path.dirname(scp), file.strip())
ok = True
for word in trans.transcriptions[
HTK_transcription.WORD][id]:
if not dic.word_in_dict(word):
print("%s skipped, because has missing word %s" %
(file.strip(), word))
ok = False
break
if ok:
print(file.strip(), file=scp_desc)
real_trans.transcriptions[real_trans.WORD][
id] = trans.transcriptions[real_trans.WORD][id]
real_trans.write_mlf(self.training_word_mlf,
target=HTK_transcription.WORD)
self.expand_word_transcription()
def __init__(self, htk_config, name, model, scp, dictionary,
language_model):
if not name.startswith('/'):
name = os.path.join(os.getcwd(), name)
if htk_config.num_speaker_chars < 0:
htk_config.num_speaker_chars = 3
self.name = name
if os.path.exists(name):
shutil.rmtree(name, ignore_errors=True)
os.mkdir(name)
self.a_id = 0
self.xforms_dir = os.path.join(name, 'xforms%d' % self.a_id)
os.mkdir(self.xforms_dir)
self.classes_dir = os.path.join(name, 'classes%d' % self.a_id)
os.mkdir(self.classes_dir)
self.model = model
self.split_scp_models = []
if '?' in scp:
self.scp = None
num_scp_speaker_chars = 1
while '?' * (num_scp_speaker_chars + 1) in scp:
num_scp_speaker_chars += 1
s_index = scp.find('?' * num_scp_speaker_chars)
speakers = [
s[s_index:s_index + num_scp_speaker_chars]
for s in glob.iglob(scp)
]
for s in speakers:
real_scp = os.path.join(name, '%s_list.scp' % s)
with open(real_scp, 'w') as scp_desc:
for line in open(
scp.replace('?' * num_scp_speaker_chars, s)):
print(os.path.join(os.path.dirname(scp), line.strip()),
file=scp_desc)
self.split_scp_models.append(
(s, real_scp, model.replace('?' * num_scp_speaker_chars,
s)))
else:
self.scp = os.path.join(name, 'list.scp')
with open(self.scp, 'w') as scp_desc:
for line in open(scp):
print(os.path.join(os.path.dirname(scp), line.strip()),
file=scp_desc)
self.dict = dictionary
d = HTK_dictionary()
d.read_dict(dictionary)
self.dict = os.path.join(name, 'dict.hdecode')
d.write_dict(self.dict, False)
d = HTK_dictionary()
d.read_dict(htk_config.adap_align_dict)
self.adap_align_dict = os.path.join(name, 'dict.hvite')
d.write_dict(self.adap_align_dict, True)
self.language_model = language_model
self.htk_config = htk_config
self.adaptations = []
self.adap_num_speaker_chars = None
self.id = 0
System.set_log_dir(os.path.basename(name))
def __init__(self, htk_config, name, model, scp, dictionary, language_model):
if not name.startswith('/'):
name = os.path.join(os.getcwd(), name)
if htk_config.num_speaker_chars < 0:
htk_config.num_speaker_chars = 3
self.name = name
if os.path.exists(name):
shutil.rmtree(name,ignore_errors=True)
os.mkdir(name)
self.a_id = 0
self.xforms_dir = os.path.join(name,'xforms%d'%self.a_id)
os.mkdir(self.xforms_dir)
self.classes_dir = os.path.join(name,'classes%d'%self.a_id)
os.mkdir(self.classes_dir)
self.model = model
self.split_scp_models = []
if '?' in scp:
self.scp = None
num_scp_speaker_chars = 1
while '?' * (num_scp_speaker_chars + 1) in scp:
num_scp_speaker_chars += 1
s_index = scp.find('?' * num_scp_speaker_chars)
speakers = [s[s_index:s_index+num_scp_speaker_chars] for s in glob.iglob(scp)]
for s in speakers:
real_scp = os.path.join(name,'%s_list.scp'%s)
with open(real_scp, 'w') as scp_desc:
for line in open(scp.replace('?' * num_scp_speaker_chars, s)):
print(os.path.join(os.path.dirname(scp), line.strip()),file=scp_desc)
self.split_scp_models.append(
(s,real_scp,model.replace('?' * num_scp_speaker_chars, s))
)
else:
self.scp = os.path.join(name,'list.scp')
with open(self.scp, 'w') as scp_desc:
for line in open(scp):
print(os.path.join(os.path.dirname(scp), line.strip()),file=scp_desc)
self.dict = dictionary
d = HTK_dictionary()
d.read_dict(dictionary)
self.dict = os.path.join(name, 'dict.hdecode')
d.write_dict(self.dict,False)
d = HTK_dictionary()
d.read_dict(htk_config.adap_align_dict)
self.adap_align_dict = os.path.join(name, 'dict.hvite')
d.write_dict(self.adap_align_dict,True)
self.language_model = language_model
self.htk_config = htk_config
self.adaptations = []
self.adap_num_speaker_chars = None
self.id = 0
System.set_log_dir(os.path.basename(name))
