def test(params):
from io_modules.encodings import Encodings
from io_modules.dataset import LTSDataset
encodings = Encodings()
encodings.load(params.model + '.encodings')
testset = LTSDataset(params.input)
from models.g2p import G2P
model = G2P(encodings)
model.load(params.model + '-bestAcc.network')
f = open(params.output, 'w')
correct = 0
last_proc = 0
index = 0
for entry in testset.entries:
index += 1
curr_proc = int(index * 100 / len(testset.entries))
if curr_proc % 5 == 0:
while last_proc < curr_proc:
last_proc += 5
sys.stdout.write(str(last_proc) + ' ')
sys.stdout.flush()
p_transcription = model.transcribe(entry.word)
ok = False
if p_transcription == entry.transcription:
correct += 1
ok = True
f.write(entry.word + '\t')
for phon in entry.transcription:
f.write(phon + ' ')
f.write('\t')
for phon in p_transcription:
f.write(phon + ' ')
if not ok:
f.write('***')
f.write('\n')
f.close()
sys.stdout.write('done\n')
sys.stdout.write('Word level accuracy is ' +
str(float(correct) / len(testset.entries)) + '\n')
def convert(params):
from io_modules.encodings import Encodings
encodings = Encodings()
encodings.load(params.model + '.encodings')
from models.g2p import G2P
model = G2P(encodings)
model.load(params.model + '-bestAcc.network')
from os import listdir
from os.path import isfile, join
files_tmp = [
f for f in listdir(params.input)
if isfile(join(params.input, f)) and f.find('txt') > -1
]
for file in files_tmp:
print(file)
with open(join(params.input, file), 'r') as f:
line = f.readline()
with open(join(params.input, file), 'w') as w:
phones = []
for word in line.split(' '):
phones.extend(model.transcribe(word))
for phone in phones:
w.write(phone + ' ')
def train(params):
from io_modules.encodings import Encodings
from io_modules.dataset import LTSDataset
sys.stdout.write('Loading datasets...\n')
trainset = LTSDataset(params.train_file)
devset = LTSDataset(params.dev_file)
sys.stdout.write('Trainset has ' + str(len(trainset.entries)) +
' entries\n')
sys.stdout.write('Devset has ' + str(len(devset.entries)) + ' entries\n')
encodings = Encodings()
encodings.update_encodings(trainset)
sys.stdout.write('Found ' + str(len(encodings.char2int)) + ' characters\n')
sys.stdout.write('Found ' + str(len(encodings.phoneme2int)) +
' phonemes\n')
from trainers.g2p import G2PTrainer
from models.g2p import G2P
model = G2P(encodings)
trainer = G2PTrainer()
trainer.start_training(model, encodings, trainset, devset,
params.model_base, params.batch_size,
params.patience)
elif not params.txt_file:
print("Input file is mandatory")
elif not params.output_file:
print("Output file is mandatory")
memory = int(params.memory)
# for compatibility we have to add this paramater
params.learning_rate = 0.0001
dynet_config.set(mem=memory, random_seed=9)
if params.gpu:
dynet_config.set_gpu()
if params.g2p is not None:
from models.g2p import G2P
from io_modules.encodings import Encodings
g2p_encodings = Encodings()
g2p_encodings.load(params.g2p + '.encodings')
g2p = G2P(g2p_encodings)
g2p.load(params.g2p + '-bestAcc.network')
if exists(params.g2p + '.lexicon'):
g2p.load_lexicon(params.g2p + '.lexicon')
else:
g2p = None
synthesize(params.speaker,
params.txt_file,
params.output_file,
params,
g2p=g2p)
def phase_1_prepare_corpus(params):
from os import listdir
from os.path import isfile, join
from os.path import exists
train_files_tmp = [
f for f in listdir(params.train_folder)
if isfile(join(params.train_folder, f))
]
if params.dev_folder is not None:
dev_files_tmp = [
f for f in listdir(params.dev_folder)
if isfile(join(params.dev_folder, f))
]
else:
dev_files_tmp = []
if params.g2p is not None:
from models.g2p import G2P
from io_modules.encodings import Encodings
g2p_encodings = Encodings()
g2p_encodings.load(params.g2p + '.encodings')
g2p = G2P(g2p_encodings)
g2p.load(params.g2p + '-bestAcc.network')
if exists(params.g2p + '.lexicon'):
g2p.load_lexicon(params.g2p + '.lexicon')
else:
g2p = None
sys.stdout.write("Scanning training files...")
sys.stdout.flush()
final_list = []
for file in train_files_tmp:
base_name = file[:-4]
lab_name = base_name + '.txt'
wav_name = base_name + '.wav'
if exists(join(params.train_folder, lab_name)) and exists(
join(params.train_folder, wav_name)):
if base_name not in final_list:
final_list.append(base_name)
train_files = final_list
sys.stdout.write(" found " + str(len(train_files)) +
" valid training files\n")
sys.stdout.write("Scanning development files...")
sys.stdout.flush()
final_list = []
for file in dev_files_tmp:
base_name = file[:-4]
lab_name = base_name + '.txt'
wav_name = base_name + '.wav'
if exists(join(params.dev_folder, lab_name)) and exists(
join(params.dev_folder, wav_name)):
if base_name not in final_list:
final_list.append(base_name)
dev_files = final_list
sys.stdout.write(" found " + str(len(dev_files)) +
" valid development files\n")
from io_modules.dataset import DatasetIO
from io_modules.vocoder import MelVocoder
from shutil import copyfile
dio = DatasetIO()
vocoder = MelVocoder()
base_folder = params.train_folder
total_files = 0
for index in range(len(train_files)):
total_files += 1
sys.stdout.write("\r\tprocessing file " + str(index + 1) + "/" +
str(len(train_files)))
sys.stdout.flush()
base_name = train_files[index]
txt_name = base_name + '.txt'
wav_name = base_name + '.wav'
spc_name = base_name + '.png'
lab_name = base_name + '.lab'
tgt_txt_name = txt_name
tgt_spc_name = spc_name
tgt_lab_name = lab_name
if params.prefix is not None:
tgt_txt_name = params.prefix + "_{:05d}".format(
total_files) + '.txt'
tgt_spc_name = params.prefix + "_{:05d}".format(
total_files) + '.png'
tgt_lab_name = params.prefix + "_{:05d}".format(
total_files) + '.lab'
# LAB - copy or create
if exists(join(base_folder, lab_name)):
copyfile(join(base_folder, lab_name),
join('data/processed/train', tgt_lab_name))
else:
create_lab_file(join(base_folder, txt_name),
join('data/processed/train', tgt_lab_name),
speaker_name=params.speaker,
g2p=g2p)
# TXT
copyfile(join(base_folder, txt_name),
join('data/processed/train', tgt_txt_name))
# WAVE
data, sample_rate = dio.read_wave(
join(base_folder, wav_name),
sample_rate=params.target_sample_rate)
mgc = vocoder.melspectrogram(data,
sample_rate=params.target_sample_rate,
num_mels=params.mgc_order)
# SPECT
render_spectrogram(mgc, join('data/processed/train', tgt_spc_name))
if params.prefix is None:
dio.write_wave(
join('data/processed/train', base_name + '.orig.wav'),
data, sample_rate)
array2file(mgc, join('data/processed/train',
base_name + '.mgc'))
else:
tgt_wav_name = params.prefix + "_{:05d}".format(
total_files) + '.orig.wav'
tgt_mgc_name = params.prefix + "_{:05d}".format(
total_files) + '.mgc'
dio.write_wave(join('data/processed/train', tgt_wav_name),
data, sample_rate)
array2file(mgc, join('data/processed/train', tgt_mgc_name))
sys.stdout.write('\n')
base_folder = params.dev_folder
for index in range(len(dev_files)):
total_files += 1
sys.stdout.write("\r\tprocessing file " + str(index + 1) + "/" +
str(len(dev_files)))
sys.stdout.flush()
base_name = dev_files[index]
txt_name = base_name + '.txt'
wav_name = base_name + '.wav'
spc_name = base_name + '.png'
lab_name = base_name + '.lab'
tgt_txt_name = txt_name
tgt_spc_name = spc_name
tgt_lab_name = lab_name
if params.prefix is not None:
tgt_txt_name = params.prefix + "_{:05d}".format(
total_files) + '.txt'
tgt_spc_name = params.prefix + "_{:05d}".format(
total_files) + '.png'
tgt_lab_name = params.prefix + "_{:05d}".format(
total_files) + '.lab'
# LAB - copy or create
if exists(join(base_folder, lab_name)):
copyfile(join(base_folder, lab_name),
join('data/processed/dev', tgt_lab_name))
else:
create_lab_file(join(base_folder, txt_name),
join('data/processed/dev', tgt_lab_name),
speaker_name=params.speaker,
g2p=g2p)
# TXT
copyfile(join(base_folder, txt_name),
join('data/processed/dev', tgt_txt_name))
# WAVE
data, sample_rate = dio.read_wave(
join(base_folder, wav_name),
sample_rate=params.target_sample_rate)
mgc = vocoder.melspectrogram(data,
sample_rate=params.target_sample_rate,
num_mels=params.mgc_order)
# SPECT
render_spectrogram(mgc, join('data/processed/dev', tgt_spc_name))
if params.prefix is None:
dio.write_wave(
join('data/processed/dev', base_name + '.orig.wav'), data,
sample_rate)
array2file(mgc, join('data/processed/dev', base_name + '.mgc'))
else:
tgt_wav_name = params.prefix + "_{:05d}".format(
total_files) + '.orig.wav'
tgt_mgc_name = params.prefix + "_{:05d}".format(
total_files) + '.mgc'
dio.write_wave(join('data/processed/dev', tgt_wav_name), data,
sample_rate)
array2file(mgc, join('data/processed/dev', tgt_mgc_name))
sys.stdout.write('\n')