Skip to content

Commit

Permalink
Merge pull request #186 from tsurumeso/develop
Browse files Browse the repository at this point in the history 
Refactor
  • Loading branch information
@tsurumeso
tsurumeso authored Jul 14, 2024
2 parents d216e56 + d40dbd6 commit 4cb4604
Show file tree
Hide file tree
Showing 2 changed files with 52 additions and 26 deletions.
48 changes: 27 additions & 21 deletions convert.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -35,29 +35,34 @@ def main():
model.to(device)
print('done')

train_filelist, val_filelist = dataset.train_val_split(
cache_dir = 'sr{}_hl{}_nf{}'.format(args.sr, args.hop_length, args.n_fft)
filelist = dataset.raw_data_split(
dataset_dir=args.dataset,
split_mode=args.split_mode,
val_rate=0,
vocals=False
split_mode=args.split_mode
)

filelist = train_filelist + val_filelist
cache_dir = 'sr{}_hl{}_nf{}'.format(args.sr, args.hop_length, args.n_fft)

for mix_path, inst_path in filelist:
X_basename = os.path.splitext(os.path.basename(mix_path))[0]
y_basename = os.path.splitext(os.path.basename(inst_path))[0]
pv_basename = X_basename + '_PseudoVocals'
pi_basename = X_basename + '_PseudoInstruments'

print('converting {}...'.format(X_basename))

X_cache_dir = os.path.join(os.path.dirname(mix_path), cache_dir)
y_cache_dir = os.path.join(os.path.dirname(inst_path), cache_dir)
v_dir = os.path.join(os.path.split(os.path.dirname(inst_path))[0], 'vocals')
v_cache_dir = os.path.join(v_dir, cache_dir)
X_dir = os.path.dirname(mix_path)
y_dir = os.path.dirname(inst_path)
pv_dir = os.path.join(os.path.split(y_dir)[0], 'pseudo_vocals')
pi_dir = os.path.join(os.path.split(y_dir)[0], 'pseudo_instruments')

X_cache_dir = os.path.join(X_dir, cache_dir)
y_cache_dir = os.path.join(y_dir, cache_dir)
pv_cache_dir = os.path.join(pv_dir, cache_dir)
pi_cache_dir = os.path.join(pi_dir, cache_dir)

os.makedirs(X_cache_dir, exist_ok=True)
os.makedirs(y_cache_dir, exist_ok=True)
os.makedirs(v_cache_dir, exist_ok=True)
os.makedirs(pv_cache_dir, exist_ok=True)
os.makedirs(pi_cache_dir, exist_ok=True)

X, sr = librosa.load(
mix_path, sr=args.sr, mono=False, dtype=np.float32, res_type='kaiser_fast')
Expand All @@ -73,19 +78,20 @@ def main():
y = spec_utils.wave_to_spectrogram(y, args.hop_length, args.n_fft)

sp = inference.Separator(model, device, args.batchsize, args.cropsize)
_, v_spec = sp.separate_tta(X - y)
# a_spec, _ = sp.separate_tta(X - y)
_, pv = sp.separate_tta(X - y)
# pa, pv = sp.separate_tta(X - y)

# pi = y + pa

# print('inverse stft of pseudo instruments...', end=' ')
# pseudo_inst = y + a_spec
# print('done')
# wave = spec_utils.spectrogram_to_wave(pv, hop_length=args.hop_length)
sf.write('{}/{}.wav'.format(pv_dir, pv_basename), [0], sr)
# wave = spec_utils.spectrogram_to_wave(pi, hop_length=args.hop_length)
sf.write('{}/{}.wav'.format(pi_dir, pi_basename), [0], sr)

np.save('{}/{}.npy'.format(X_cache_dir, X_basename), X.transpose(2, 0, 1))
np.save('{}/{}.npy'.format(y_cache_dir, y_basename), y.transpose(2, 0, 1))

wave = spec_utils.spectrogram_to_wave(v_spec, hop_length=args.hop_length)
sf.write('{}/{}_Vocals.wav'.format(v_dir, X_basename), wave.T, sr)
np.save('{}/{}_Vocals.npy'.format(v_cache_dir, X_basename), v_spec.transpose(2, 0, 1))
np.save('{}/{}.npy'.format(pv_cache_dir, pv_basename), pv.transpose(2, 0, 1))
# np.save('{}/{}.npy'.format(pi_cache_dir, pi_basename), pi.transpose(2, 0, 1))


if __name__ == '__main__':
Expand Down
30 changes: 25 additions & 5 deletions lib/dataset.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -176,12 +176,12 @@ def make_pair(X_dir, y_dir, v_dir=None):
return filelist


def train_val_split(dataset_dir, split_mode, val_rate, val_filelist=[], vocals=True):
def train_val_split(dataset_dir, split_mode, val_rate, val_filelist=[]):
if split_mode == 'random':
filelist = make_pair(
os.path.join(dataset_dir, 'mixtures'),
os.path.join(dataset_dir, 'instruments'),
os.path.join(dataset_dir, 'vocals') if vocals else None
os.path.join(dataset_dir, 'pseudo_vocals')
)

random.shuffle(filelist)
Expand All @@ -198,22 +198,42 @@ def train_val_split(dataset_dir, split_mode, val_rate, val_filelist=[], vocals=T
elif split_mode == 'subdirs':
if len(val_filelist) != 0:
raise ValueError('`val_filelist` option is not available with `subdirs` mode')

train_filelist = make_pair(
os.path.join(dataset_dir, 'training/mixtures'),
os.path.join(dataset_dir, 'training/instruments'),
os.path.join(dataset_dir, 'training/vocals') if vocals else None
os.path.join(dataset_dir, 'training/pseudo_vocals')
)

val_filelist = make_pair(
os.path.join(dataset_dir, 'validation/mixtures'),
os.path.join(dataset_dir, 'validation/instruments'),
os.path.join(dataset_dir, 'validation/vocals') if vocals else None
os.path.join(dataset_dir, 'validation/pseudo_vocals')
)

return train_filelist, val_filelist


def raw_data_split(dataset_dir, split_mode):
if split_mode == 'random':
filelist = make_pair(
os.path.join(dataset_dir, 'mixtures'),
os.path.join(dataset_dir, 'instruments'),
)
elif split_mode == 'subdirs':
train_filelist = make_pair(
os.path.join(dataset_dir, 'training/mixtures'),
os.path.join(dataset_dir, 'training/instruments'),
)
val_filelist = make_pair(
os.path.join(dataset_dir, 'validation/mixtures'),
os.path.join(dataset_dir, 'validation/instruments'),
)
filelist = train_filelist + val_filelist

return filelist


def make_padding(width, cropsize, offset):
left = offset
roi_size = cropsize - offset * 2
Expand Down

0 comments on commit 4cb4604

Please sign in to comment.