def main(config):
use_cuda = config['use_gpu']
device = torch.device("cuda" if use_cuda == 1 else "cpu")
model = Multi_modal(config['input_dim'], config['vocab_len'],
config['embedding_dim'], config['hidden_dim_AE'],
config['hidden_dim_PE'])
model = model.to(device)
optimizer = ScheduledOptim(
optim.Adam(filter(lambda p: p.requires_grad, model.parameters()),
betas=(0.9, 0.98),
eps=1e-09,
weight_decay=1e-4,
amsgrad=True), config['n_warmup_steps'])
criterion = nn.CrossEntropyLoss()
### Data related
dataset_train = SpeechDataGenerator(manifest=args.training_filepath)
dataloader_train = DataLoader(dataset_train,
batch_size=config['batch_size'],
shuffle=True,
collate_fn=speech_collate)
dataset_test = SpeechDataGenerator(manifest=args.testing_filepath)
dataloader_test = DataLoader(dataset_test,
batch_size=config['batch_size'],
shuffle=True,
collate_fn=speech_collate)
if not os.path.exists(args.save_modelpath):
os.makedirs(args.save_modelpath)
best_acc = 0
best_loss = np.inf
best_epoch = -1
for epoch in range(1, config['num_epochs'] + 1):
train_loss, train_acc = train(model, dataloader_train, epoch,
optimizer, device, criterion)
val_loss, val_acc = evaluation(model, dataloader_test, epoch, device,
criterion)
# Save
if val_acc > best_acc:
best_acc = max(val_acc, best_acc)
model_save_path = os.path.join(args.save_modelpath,
'best_check_point_' + str(epoch))
state_dict = {
'model': model.state_dict(),
'optimizer': optimizer.state_dict(),
'epoch': epoch
}
torch.save(state_dict, model_save_path)
print('best accuracy so far {}'.format(best_acc))
best_epoch = epoch + 1
elif epoch - best_epoch > 2:
optimizer.increase_delta()
best_epoch = epoch + 1
print('best accuracy so far {}'.format(best_acc))
def main(config,args):
use_cuda = config['use_gpu']
device = torch.device("cuda" if use_cuda==1 else "cpu")
model = Siamese()
model = model.to(device)
rec_loss = nn.L1Loss()
cosine_loss = nn.CosineSimilarity(dim=1)
optimizer = optim.Adam(model.parameters(), lr=0.001, weight_decay=0.0, betas=(0.9, 0.98), eps=1e-9)
dataset_train = SpeechDataGenerator(args.clean_file,args.noisy_file,batch_s=100)
dataloader_train = DataLoader(dataset_train, batch_size=1,shuffle=True,collate_fn=speech_collate)
for epoch in range(1, config['num_epochs'] + 1):
train_loss=train(model,dataloader_train,epoch,optimizer,device,rec_loss,cosine_loss)
def main(args):
### Data related
dataset_train = SpeechDataGenerator(manifest=args.training_filepath,
mode='train')
dataloader_train = DataLoader(dataset_train,
batch_size=args.batch_size,
shuffle=True,
collate_fn=speech_collate)
dataset_val = SpeechDataGenerator(manifest=args.validation_filepath,
mode='train')
dataloader_val = DataLoader(dataset_val,
batch_size=args.batch_size,
shuffle=True,
collate_fn=speech_collate)
#dataset_test = SpeechDataGenerator(manifest=args.testing_filepath,mode='test')
#dataloader_test = DataLoader(dataset_test, batch_size=args.batch_size,shuffle=True,collate_fn=speech_collate)
## Model related
use_cuda = torch.cuda.is_available()
device = torch.device("cuda" if use_cuda else "cpu")
model = RawNet(args.input_dim, args.num_classes).to(device)
optimizer = optim.Adam(model.parameters(),
lr=0.001,
weight_decay=0.0,
betas=(0.9, 0.98),
eps=1e-9)
ce_loss = nn.CrossEntropyLoss()
g2e_loss = GE2ELoss(device)
for epoch in range(args.num_epochs):
train(model, dataloader_train, epoch, ce_loss, g2e_loss, optimizer,
device)
validation(model, dataloader_val, epoch, ce_loss, g2e_loss, optimizer,
device)
default='meta/training_s1_s2_s3_s4.txt')
parser.add_argument('-testing_filepath',
type=str,
default='meta/testing_s5.txt')
parser.add_argument('-input_dim', action="store_true", default=1)
parser.add_argument('-num_classes', action="store_true", default=4)
parser.add_argument('-lamda_val', action="store_true", default=0.1)
parser.add_argument('-batch_size', action="store_true", default=64)
parser.add_argument('-use_gpu', action="store_true", default=True)
parser.add_argument('-num_epochs', action="store_true", default=100)
args = parser.parse_args()
### Data related
dataset_test = SpeechDataGenerator(manifest=args.testing_filepath, mode='test')
dataloader_test = DataLoader(dataset_test,
batch_size=args.batch_size,
shuffle=True,
collate_fn=speech_collate)
## Model related
use_cuda = torch.cuda.is_available()
device = torch.device("cuda" if use_cuda else "cpu")
model = Emo_Raw_TDNN(args.input_dim, args.num_classes).to(device)
optimizer = optim.Adam(model.parameters(),
lr=0.001,
weight_decay=0.0,
betas=(0.9, 0.98),
eps=1e-9)
data_path_train = 'meta/training.txt'
data_path_test = 'meta/testing.txt'
#### Params
def speech_collate(batch):
targets = []
specs = []
for sample in batch:
specs.append(sample['spec'])
targets.append((sample['labels']))
return specs, targets
### Data related
dataset_train = SpeechDataGenerator(manifest=data_path_train, mode='train')
dataloader_train = DataLoader(dataset_train,
batch_size=32,
shuffle=True,
collate_fn=speech_collate)
dataset_test = SpeechDataGenerator(manifest=data_path_test, mode='test')
dataloader_test = DataLoader(dataset_test,
batch_size=32,
collate_fn=speech_collate)
## Model related
use_cuda = torch.cuda.is_available()
#device = torch.device("cuda" if use_cuda else "cpu")
device = torch.device("cuda:0")
model = AttnPooling(num_classes=1).to(device)
#model.load_state_dict(torch.load('model_checkpoints/check_point_old')['model'])
data_path_train = 'meta/training.txt'
data_path_test = 'meta/testing.txt'
#### Params
def speech_collate(batch):
targets = []
specs = []
for sample in batch:
specs.append(sample['audio_wav'])
targets.append((sample['labels']))
return specs, targets
### Data related
dataset_train = SpeechDataGenerator(manifest=data_path_train)
dataloader_train = DataLoader(dataset_train,
batch_size=150,
num_workers=12,
shuffle=True,
collate_fn=speech_collate)
dataset_test = SpeechDataGenerator(manifest=data_path_test)
dataloader_test = DataLoader(dataset_test,
batch_size=150,
num_workers=12,
shuffle=True,
collate_fn=speech_collate)
## Model related
use_cuda = torch.cuda.is_available()
device = torch.device("cuda" if use_cuda else "cpu")
parser.add_argument('-training_filepath',
type=str,
default='meta/training.txt')
parser.add_argument('-testing_filepath', type=str, default='meta/testing.txt')
parser.add_argument('-input_feat_dim', action="store_true", default=39)
parser.add_argument('-num_phones', action="store_true", default=86)
parser.add_argument('-num_heads', action="store_true", default=13)
parser.add_argument('-num_layers', action="store_true", default=12)
parser.add_argument('-lamda_val', action="store_true", default=0.1)
parser.add_argument('-batch_size', action="store_true", default=32)
parser.add_argument('-use_gpu', action="store_true", default=True)
parser.add_argument('-num_epochs', action="store_true", default=100)
args = parser.parse_args()
### Data related
dataset_train = SpeechDataGenerator(manifest=args.training_filepath)
dataloader_train = DataLoader(dataset_train,
batch_size=args.batch_size,
shuffle=True,
collate_fn=speech_collate)
dataset_test = SpeechDataGenerator(manifest=args.testing_filepath)
dataloader_test = DataLoader(dataset_test,
batch_size=args.batch_size,
shuffle=True,
collate_fn=speech_collate)
## Model related
use_cuda = torch.cuda.is_available()
device = torch.device("cuda" if use_cuda else "cpu")
model = BERTphone(args.input_feat_dim, args.num_phones, args.num_heads,
default='meta/training.txt')
parser.add_argument('-testing_filepath', type=str, default='meta/testing.txt')
parser.add_argument('-validation_filepath',
type=str,
default='meta/validation.txt')
parser.add_argument('-input_dim', action="store_true", default=40)
parser.add_argument('-num_classes', action="store_true", default=8)
parser.add_argument('-lamda_val', action="store_true", default=0.1)
parser.add_argument('-batch_size', action="store_true", default=128)
parser.add_argument('-use_gpu', action="store_true", default=True)
parser.add_argument('-num_epochs', action="store_true", default=100)
args = parser.parse_args()
### Data related
dataset_train = SpeechDataGenerator(manifest=args.training_filepath,
mode='train')
dataloader_train = DataLoader(dataset_train,
batch_size=args.batch_size,
shuffle=True,
collate_fn=speech_collate)
dataset_val = SpeechDataGenerator(manifest=args.validation_filepath,
mode='train')
dataloader_val = DataLoader(dataset_train,
batch_size=args.batch_size,
shuffle=True,
collate_fn=speech_collate)
dataset_test = SpeechDataGenerator(manifest=args.testing_filepath, mode='test')
dataloader_test = DataLoader(dataset_test,
batch_size=args.batch_size,
