def __getitem__(self, idx):
i, src_fn = self.data[idx]
src_wav = get_wav(src_fn)
src_mel = self._to_mel(src_wav)
if self.indices is None:
return src_mel, src_mel
_, tgt_fn = self.data[random.choice(self.indices[i])]
tgt_wav = get_wav(tgt_fn)
tgt_mel = self._to_mel(tgt_wav)
return src_mel, tgt_mel
def f(fpath):
w, q = get_wav(fpath)
fname = os.path.basename(fpath).replace('wav', 'npy')
if not os.path.exists("/data/private/speech/vctk/wavs"):
os.makedirs("/data/private/speech/vctk/wavs")
if not os.path.exists("/data/private/speech/vctk/qts"):
os.makedirs("/data/private/speech/vctk/qts")
np.save("/data/private/speech/vctk/wavs/{}".format(fname), w)
np.save("/data/private/speech/vctk/qts/{}".format(fname), q)
def process_one(fn, data_type, wav2vec=None, wav2mel=None):
if data_type == DataType.NORMAL:
wav = get_wav(fn)
return wav
elif data_type == DataType.WORLD:
f0, sp, ap = get_world_features(fn)
return f0, sp, ap
elif data_type == DataType.WAV2VEC:
feat, mel = get_wav2vec_features(fn, wav2vec, wav2mel)
return feat, mel
else:
raise ValueError('Invalid value: type of data_type must be DataType')
def __getitem__(self, idx):
fn = self.data[idx]
wav = get_wav(fn)
mel = self._to_mel(wav)
return wav, mel
def f(fpath):
w, q = get_wav(fpath)
fname = os.path.basename(fpath).replace('wav', 'npy')
np.save(hp.transformed_data_wav + "/{}".format(fname), w)
np.save(hp.transformed_data_qts + "/{}".format(fname), q)
parser.add_argument('--src_path', type=str, required=True)
parser.add_argument('--tgt_path', type=str, required=True)
parser.add_argument('--config_path', type=str, required=True)
parser.add_argument('--ckpt_path', type=str, required=True)
parser.add_argument('--output_dir', type=str, default='./outputs')
args = parser.parse_args()
params = get_config(args.config_path)
output_dir = pathlib.Path(args.output_dir) / args.ckpt_path.split('/')[-2]
if not output_dir.exists():
output_dir.mkdir(parents=True)
print('Build model')
model = module_from_config(params)
model = model.load_from_checkpoint(args.ckpt_path)
model.freeze()
print(model.hparams)
print('Inference')
wav = model(args.src_path, args.tgt_path)
print('Saving')
src_wav, tgt_wav = get_wav(args.src_path), get_wav(args.tgt_path)
save_sample(str(output_dir / 'src.wav'), src_wav)
save_sample(str(output_dir / 'tgt.wav'), tgt_wav)
save_sample(str(output_dir / 'gen.wav'), wav)
print('End')