def main():
pprint(vars(arguments))
arguments.output.mkdir(exist_ok=True)
save_arguments(arguments, arguments.output / 'arguments.json')
path_feature1 = {
Path(p).stem: Path(p)
for p in glob.glob(arguments.input_feature_glob1)
}
path_feature2 = {
Path(p).stem: Path(p)
for p in glob.glob(arguments.input_feature_glob2)
}
path_indexes = {
Path(p).stem: Path(p)
for p in glob.glob(arguments.input_indexes)
}
fn_both_list = set(path_feature1.keys()) & set(path_indexes.keys())
pool = multiprocessing.Pool()
generate = partial(
generate_aligned_wave,
sampling_rate=arguments.sampling_rate,
frame_period=arguments.frame_period,
alpha=arguments.alpha,
)
it = pool.imap(generate,
((path_feature1[fn], path_feature2[fn], path_indexes[fn])
for fn in fn_both_list))
list(tqdm.tqdm(it, total=len(path_feature1)))
def main():
arguments.output.mkdir(exist_ok=True)
save_arguments(arguments, arguments.output / 'arguments.json')
paths = [Path(p) for p in glob.glob(arguments.input_glob)]
pool = multiprocessing.Pool()
it = pool.imap(generate_feature, paths)
list(tqdm.tqdm(it, total=len(paths)))
def main():
save_arguments(arguments, output / 'arguments.json')
# f0 converter
if f0_trans_model_dir is not None:
model = _get_predictor_model_path(f0_trans_model_dir,
f0_trans_model_iteration)
f0_converter = AcousticConverter(create_config(f0_trans_config),
model,
gpu=gpu)
elif input_statistics is not None:
f0_converter = F0Converter(input_statistics=input_statistics,
target_statistics=target_statistics)
else:
f0_converter = None
# acoustic converter
config = create_config(voice_changer_config)
model = _get_predictor_model_path(voice_changer_model_dir,
voice_changer_model_iteration)
acoustic_converter = AcousticConverter(
config,
model,
gpu=gpu,
f0_converter=f0_converter,
out_sampling_rate=arguments.out_sampling_rate,
)
print(f'Loaded acoustic converter model "{model}"')
# super resolution
sr_config = create_sr_config(super_resolution_config)
super_resolution = SuperResolution(sr_config,
super_resolution_model,
gpu=gpu)
print(f'Loaded super resolution model "{super_resolution_model}"')
# dataset's test
if not disable_dataset_test:
input_paths = list(
sorted(
[Path(p) for p in glob.glob(str(config.dataset.input_glob))]))
numpy.random.RandomState(config.dataset.seed).shuffle(input_paths)
paths_test = input_paths[-config.dataset.num_test:]
else:
paths_test = []
# test data
if test_wave_dir is not None:
paths_test += list(test_wave_dir.glob('*.wav'))
process_partial = partial(process,
acoustic_converter=acoustic_converter,
super_resolution=super_resolution)
if gpu is None:
list(multiprocessing.Pool().map(process_partial, paths_test))
else:
list(map(process_partial, paths_test))
def main():
arguments.output.mkdir(exist_ok=True)
save_arguments(arguments, arguments.output / 'arguments.json')
paths1 = [Path(p) for p in sorted(glob.glob(arguments.input_wave_glob1))]
paths2 = [Path(p) for p in sorted(glob.glob(arguments.input_wave_glob2))]
assert len(paths1) == len(paths2)
pool = multiprocessing.Pool()
it = pool.imap(generate_align_indexes, zip(paths1, paths2))
list(tqdm.tqdm(it, total=len(paths1)))
def main():
arguments.output.mkdir(exist_ok=True)
save_arguments(arguments, arguments.output / 'arguments.json')
paths1 = {Path(p).stem: Path(p) for p in glob.glob(arguments.input_glob1)}
paths2 = {Path(p).stem: Path(p) for p in glob.glob(arguments.input_glob2)}
fn_both_list = set(paths1.keys()) & set(paths2.keys())
pool = multiprocessing.Pool()
it = pool.imap(generate_align_indexes,
((paths1[fn], paths2[fn]) for fn in fn_both_list))
list(tqdm.tqdm(it, total=len(fn_both_list)))
def main():
arguments.output.mkdir(exist_ok=True)
save_arguments(arguments, arguments.output / 'arguments.json')
config = create_config(arguments.vc_config)
acoustic_converter = AcousticConverter(config,
arguments.vc_model,
gpu=arguments.gpu)
paths = [Path(p) for p in glob.glob(arguments.input_glob)]
pool = multiprocessing.Pool()
it = pool.imap(
partial(convert_feature, acoustic_converter=acoustic_converter), paths)
list(tqdm.tqdm(it, total=len(paths)))
def main():
output.mkdir(exist_ok=True)
save_arguments(arguments, output / 'arguments.json')
paths1 = {Path(p).stem: Path(p) for p in glob.glob(str(input_glob))}
paths2 = {Path(p).stem: Path(p) for p in glob.glob(str(target_glob))}
paths_index = {
Path(p).stem: Path(p)
for p in glob.glob(str(aligned_index_glob))
}
fn_both_list = set(paths1.keys()) & set(paths2.keys()) & set(
paths_index.keys())
pool = multiprocessing.Pool()
it = pool.imap(generate_file, ((paths1[fn], paths2[fn], paths_index[fn])
for fn in fn_both_list))
list(tqdm.tqdm(it, total=len(fn_both_list)))
def main():
save_arguments(arguments, output_dir / 'arguments.json')
# f0 converter
if input_statistics is not None:
f0_converter = F0Converter(input_statistics=input_statistics,
target_statistics=target_statistics)
else:
f0_converter = None
# acoustic converter
config = create_config(config_path)
model = _get_predictor_model_path(model_dir, model_iteration)
acoustic_converter = AcousticConverter(
config,
model,
gpu=gpu,
f0_converter=f0_converter,
out_sampling_rate=output_sampling_rate,
)
print(f'Loaded acoustic converter model "{model}"')
# dataset test
if not disable_dataset_test:
input_paths = list(
sorted(
[Path(p) for p in glob.glob(str(config.dataset.input_glob))]))
numpy.random.RandomState(config.dataset.seed).shuffle(input_paths)
paths_test = input_paths[-config.dataset.num_test:]
else:
paths_test = []
# additional test
if test_wave_dir is not None:
paths_test += list(test_wave_dir.glob('*.wav'))
process_partial = partial(process, acoustic_converter=acoustic_converter)
if gpu is None:
list(multiprocessing.Pool().map(process_partial, paths_test))
else:
list(map(process_partial, paths_test))