def main():
args = parse_args()
vocoder = MelVocoder(args.load_path)
args.save_path.mkdir(exist_ok=True, parents=True)
for i, fname in tqdm(enumerate(args.folder.glob("*.wav"))):
wavname = fname.name
wav, sr = librosa.core.load(fname)
mel, _ = vocoder(torch.from_numpy(wav)[None])
recons = vocoder.inverse(mel).squeeze().cpu().numpy()
librosa.output.write_wav(args.save_path / wavname, recons, sr=sr)
def main():
args = parse_args()
vocoder = MelVocoder(args.load_path)
args.save_path.mkdir(exist_ok=True, parents=True)
for i, fname in tqdm(enumerate(args.pt_path.glob('*.pt'))):
wavname = os.path.splitext(fname.name)[0] + '.wav'
print('fname', fname)
print('wavname', wavname)
spectrogram = torch.load(fname)
if (len(spectrogram.shape) == 2):
spectrogram = spectrogram.unsqueeze(0)
reconstruction = vocoder.inverse(spectrogram).squeeze().cpu().numpy()
librosa.output.write_wav(str(args.save_path) + '/' + wavname, reconstruction, sr=22050)
def load_melgan(model_name="multi_speaker"):
"""
Exposes a MelVocoder Interface
Args:
model_name (str): Supports only 2 models, 'linda_johnson' or 'multi_speaker'
Returns:
object (MelVocoder): MelVocoder class.
Default function (___call__) converts raw audio to mel
inverse function convert mel to raw audio using MelGAN
"""
return MelVocoder(path=None, github=True, model_name=model_name)