def audio2mel(audiopaths_and_text, melpaths_and_text, args):
melpaths_and_text_list = load_filepaths_and_text(melpaths_and_text)
audiopaths_and_text_list = load_filepaths_and_text(audiopaths_and_text)
data_loader = TextMelLoader(audiopaths_and_text, args)
for i in range(len(melpaths_and_text_list)):
if i % 100 == 0:
print("done", i, "/", len(melpaths_and_text_list))
mel = data_loader.get_mel(audiopaths_and_text_list[i][0])
torch.save(mel, melpaths_and_text_list[i][0])
def get_data_loader(model_name, audiopaths_and_text, args):
if model_name == 'Tacotron2':
data_loader = TextMelLoader(audiopaths_and_text, args)
elif model_name == 'WaveGlow':
data_loader = MelAudioLoader(audiopaths_and_text, args)
else:
raise NotImplementedError(
"unknown data loader requested: {}".format(model_name))
return data_loader
def get_data_loader(model_name,
dataset_path,
audiopaths_and_text,
args,
speaker_ids=None):
if model_name == 'Tacotron2':
if speaker_ids is not None:
data_loader = TextMelLoader(dataset_path,
audiopaths_and_text,
args,
speaker_ids=speaker_ids)
else:
data_loader = TextMelLoader(dataset_path, audiopaths_and_text,
args)
elif model_name == 'WaveGlow':
data_loader = MelAudioLoader(dataset_path, audiopaths_and_text, args)
else:
raise NotImplementedError(
"unknown data loader requested: {}".format(model_name))
return data_loader
def audio2mel(dataset_path: str, audiopaths_and_text: str,
melpaths_and_text: str, args: ArgumentParser) -> None:
"""Create mel spectrograms on disk from audio files.
Args:
dataset_path (str): Path to dataset
audiopaths_and_text (str): Path to filelist with audio paths and text
melpaths_and_text (str): Path to filelist with mel paths and text
args (ArgumentParser): Namespace with arguments
"""
melpaths_and_text_list = load_filepaths_and_text(dataset_path,
melpaths_and_text)
audiopaths_and_text_list = load_filepaths_and_text(dataset_path,
audiopaths_and_text)
data_loader = TextMelLoader(dataset_path, audiopaths_and_text, args)
for i, melpath_and_text in enumerate(melpaths_and_text_list):
if i % 100 == 0:
print("done", i, "/", len(melpaths_and_text_list))
mel = data_loader.get_mel(audiopaths_and_text_list[i][0])
torch.save(mel, melpath_and_text[0])
def audio2mel(dataset_path,
audiopaths_and_text,
melpaths_and_text,
args,
use_intermed=None):
melpaths_and_text_list = \
load_filepaths_and_text(dataset_path, melpaths_and_text)
audiopaths_and_text_list = \
load_filepaths_and_text(dataset_path, audiopaths_and_text)
# n = 10
# print(f"The first {n} melpaths and text are {melpaths_and_text_list[:n]}")
# print(f"The first {n} audiopaths and text are {audiopaths_and_text_list[:n]}")
data_loader = TextMelLoader(dataset_path, audiopaths_and_text, args)
for i in range(len(melpaths_and_text_list)):
if i % 100 == 0:
print("done", i, "/", len(melpaths_and_text_list))
mel = data_loader.get_mel(audiopaths_and_text_list[i][0])
torch.save(mel, melpaths_and_text_list[i][0])