help='choices of neural network')
args = parser.parse_args()
train_transform = Compose(
[LoadMagSpectrogram(),
ComputeMelSpectrogramFromMagSpectrogram()])
if args.dataset == 'librispeech':
from datasets.libri_speech import LibriSpeech, vocab
train_dataset = ConcatDataset([
LibriSpeech(name='train-clean-100', transform=train_transform),
LibriSpeech(name='train-clean-360', transform=train_transform),
LibriSpeech(name='train-other-500', transform=train_transform)
])
else:
from datasets.bolor_speech import BolorSpeech, vocab
train_dataset = BolorSpeech(name='train', transform=train_transform)
directory = Path(args.input)
files = [f for f in directory.iterdir() if f.suffix == ".pth"]
assert (len(files) > 1)
def load_model(f):
if args.model == 'jasper':
model = TinyJasper(vocab)
elif args.model == 'w2l':
model = TinyWav2Letter(vocab)
else:
model = Speech2TextCRNN(vocab)
checkpoint = torch.load(f)
model.load_state_dict(checkpoint['state_dict'])
model.float()
dataset = MBSpeech()
elif args.dataset == 'librispeech':
from datasets.libri_speech import LibriSpeech
dataset = ConcatDataset([
LibriSpeech(name='train-clean-100'),
LibriSpeech(name='train-clean-360'),
LibriSpeech(name='train-other-500'),
LibriSpeech(name='dev-clean',)
])
elif args.dataset == 'backgroundsounds':
from datasets.background_sounds import BackgroundSounds
dataset = BackgroundSounds(is_random=False)
elif args.dataset == 'bolorspeech':
from datasets.bolor_speech import BolorSpeech
dataset = ConcatDataset([
BolorSpeech(name='train'),
BolorSpeech(name='train2'),
BolorSpeech(name='test'),
BolorSpeech(name='demo'),
BolorSpeech(name='annotation'),
BolorSpeech(name='annotation-1111')
])
else:
print("unknown dataset!")
import sys
sys.exit(1)
transform=Compose([LoadAudio(), ComputeMagSpectrogram()])
for data in tqdm(dataset):
fname = data['fname']
formatter_class=argparse.ArgumentDefaultsHelpFormatter)
parser.add_argument("--dataset",
choices=['librispeech', 'mbspeech', 'bolorspeech'],
default='bolorspeech',
help='dataset name')
args = parser.parse_args()
if args.dataset == 'mbspeech':
from datasets.mb_speech import MBSpeech
dataset = MBSpeech()
elif args.dataset == 'librispeech':
from datasets.libri_speech import LibriSpeech
dataset = ConcatDataset([
LibriSpeech(name='train-clean-100'),
LibriSpeech(name='train-clean-360'),
LibriSpeech(name='train-other-500'),
LibriSpeech(name='dev-clean', )
])
else:
from datasets.bolor_speech import BolorSpeech
dataset = ConcatDataset(
[BolorSpeech(name='train'),
BolorSpeech(name='test')])
transform = Compose([LoadAudio(), ComputeMagSpectrogram()])
for data in tqdm(dataset):
fname = data['fname']
data = transform(data)
mel_spectrogram = data['input']
np.save(fname.replace('.wav', '.npy'), mel_spectrogram)