def save_audio(mag_spec, logdir, name, stft, train=True):
magnitudes = dynamic_range_decompression(mag_spec)
magnitudes = torch.pow(magnitudes, 1.2)
magnitudes = torch.unsqueeze(magnitudes, 0)
# magnitudes = torch.t(magnitudes)
# print(magnitudes.shape)
signal = griffin_lim(magnitudes.cpu(), stft.stft_fn)
signal = signal.data.cpu().numpy()
if train:
file_name = '{}/sample_train_step_{}.wav'.format(logdir, name)
else:
file_name = '{}/sample_eval_step_{}.wav'.format(logdir, name)
if logdir[0] != '/':
file_name = "./"+file_name
print(signal.shape)
# signal = signal.astype(np.int16)
signal = signal[0]
# print(signal)
write(file_name, 22050 ,signal)
def spectral_de_normalize(self, magnitudes):
output = dynamic_range_decompression(magnitudes)
return output
def normalize_mel(mel, mean, sigma):
normalized = dynamic_range_decompression(mel)
normalized = convert_mel(normalized)
normalized = (np.log10(normalized) - mean) / sigma
return normalized
