def get_wavernn(config_path, checkpoint_path, device):
# Init model
params_wavernn = load_config(config_path)
wavernn = WaveRNN(**params_wavernn).to(device)
# Load checkpoint
state_dict = torch.load(checkpoint_path,
map_location=lambda storage, loc: storage)
wavernn.load_state_dict(state_dict)
print("Loaded WaveRNN checkpoint.")
return wavernn, params_wavernn
def get_tacotron(config_path, checkpoint_path, device):
# Init model
params_tacotron = load_config(config_path)
params_tacotron["num_chars"] = len(char_list_g2p)
tacotron = Tacotron2(**params_tacotron).to(device)
# Load checkpoint
state_dict = torch.load(checkpoint_path,
map_location=lambda storage, loc: storage)
tacotron.load_state_dict(state_dict)
print("Loaded Tacotron checkpoint.")
return tacotron, params_tacotron
def main(args):
# Load config
config_file_path = os.path.join(args.config_path, "config.yml")
config = load_config(config_file_path)
# Set number of characters
config["model"]["num_chars"] = len(char_list)
# Path manager
output_path = os.path.join(config["output_path"], config["run_name"])
path_manager = PathManager(output_path)
# Model
model = Tacotron2(**config["model"])
# Trainer
trainer = TacotronTrainer(config, path_manager, model)
trainer.train()
meta_line = executor.submit(
partial(compute_features, source_audio_path, file_name_, text,
speaker_name, out_melspecs_path, g2p, itr,
count_files))
metafile.append(meta_line)
metafile = [
metaline.result() for metaline in metafile if metaline is not None
]
print(metafile)
# Write metafile
with open(os.path.join(output_path, "metadata.txt"), "w") as final_meta:
for l in metafile:
final_meta.write(l + "\n")
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument("--dataset_path", type=str, required=True)
parser.add_argument("--output_path", type=str, required=True)
parser.add_argument("--config_path", type=str, required=True)
parser.add_argument("--num_workers", type=int, default=5)
args = parser.parse_args()
target_speakers = [
"0d358649ded3baf7f476eeb2ba44fc2cfc195824b0294fcb4a2059c4e6a2e6ab1aede4dd71f5df11fb4550d6db6ee9e45244180d9692ea897afb86cc0471caa0"
]
config = load_config(os.path.join(args.config_path, "config.yml"))
preprocess(args.dataset_path, args.output_path, target_speakers, config,
args.num_workers)