def load_data(self, split='train', load_mel_only=False):
''' Load data for training / testing'''
if split == 'train':
self.verbose('Loading source data ' +
str(self.config['dataloader']['train_set']) +
' from ' + self.config['dataloader']['data_path'])
if self.duo_feature:
self.verbose('Loading target data ' +
str(self.config['dataloader']['train_set']) +
' from ' +
self.config['dataloader']['target_path'])
elif split == 'test':
self.verbose('Loading testing data ' +
str(self.config['dataloader']['test_set']) +
' from ' + self.config['dataloader']['data_path'])
else:
raise NotImplementedError('Invalid `split` argument!')
if self.duo_feature and not load_mel_only:
setattr(self, 'dataloader', get_Dataloader(split, load='duo', use_gpu=self.paras.gpu, \
mock_config=self.config['mockingjay'], **self.config['dataloader'])) # Currently the duo feature dataloader only supports mockingjay training, no need to specify `run_mockingjay`
else:
setattr(self, 'dataloader', get_Dataloader(split, load='acoustic', use_gpu=self.paras.gpu, \
run_mockingjay=True if not load_mel_only else False, mock_config=self.config['mockingjay'], \
**self.config['dataloader'])) # specify `run_mockingjay` so dataloader will process mockingjay MAM data
def get_dataloader(args, dataloader_config):
pretrain_config = torch.load(args.ckpt, map_location='cpu')['Settings']['Config']
if 'online' in pretrain_config:
dataloader_config['online_config'] = pretrain_config['online']
if not os.path.exists(dataloader_config['data_path']):
raise RuntimeError('[run_downstream] - Data path not valid:', dataloader_config['data_path'])
print('[run_downstream] - Loading input data: ' + str(dataloader_config['train_set']) + ' from ' + dataloader_config['data_path'])
if args.task == 'speaker':
print('[run_downstream] - Loading speaker data: ' + str(dataloader_config['train_set']) + ' from ' + dataloader_config['data_path'])
else:
print('[run_downstream] - Loading phone data: ' + dataloader_config['phone_path'])
if not os.path.exists(dataloader_config['phone_path']):
raise RuntimeError('[run_downstream] - Phone path not valid:', dataloader_config['phone_path'])
if args.task == 'montreal_phone':
print('[run_downstream] - WARNING: Using a non-preset phone set! Please make sure \'data_path\' (should be: data/libri_mel160_subword5000) and \'phone_path\' (should be: data/libri_phone) are set correctly.')
print('[run_downstream] - getting train dataloader...')
train_loader = get_Dataloader(split='train', load=args.task, use_gpu=args.gpu, seed=args.seed, **dataloader_config)
print('[run_downstream] - getting dev dataloader...')
dev_loader = get_Dataloader(split='dev', load=args.task, use_gpu=args.gpu, seed=args.seed, **dataloader_config)
print('[run_downstream] - getting test dataloader...')
test_loader = get_Dataloader(split='test', load=args.task, use_gpu=args.gpu, seed=args.seed, **dataloader_config)
return train_loader, dev_loader, test_loader
def get_dataloader(args, dataloader_config):
if not os.path.exists(dataloader_config['data_path']):
raise RuntimeError('[run_downstream] - Data path not valid:',
dataloader_config['data_path'])
print('[run_downstream] - Loading input data: ' +
str(dataloader_config['train_set']) + ' from ' +
dataloader_config['data_path'])
if args.task == 'speaker':
print('[run_downstream] - Loading speaker data: ' +
str(dataloader_config['train_set']) + ' from ' +
dataloader_config['data_path'])
else:
print('[run_downstream] - Loading phone data: ' +
dataloader_config['phone_path'])
print('[run_downstream] - getting train dataloader...')
train_loader = get_Dataloader(split='train',
load=args.task,
use_gpu=args.gpu,
**dataloader_config)
print('[run_downstream] - getting dev dataloader...')
dev_loader = get_Dataloader(split='dev',
load=args.task,
use_gpu=args.gpu,
**dataloader_config)
print('[run_downstream] - getting test dataloader...')
test_loader = get_Dataloader(split='test',
load=args.task,
use_gpu=args.gpu,
**dataloader_config)
return train_loader, dev_loader, test_loader
def load_data(self):
''' Load date for training/validation'''
self.verbose('Loading data from ' + self.config['solver']['data_path'])
setattr(self, 'train_set', get_Dataloader('train', load='asr', use_gpu=self.paras.gpu, **self.config['solver']))
setattr(self, 'dev_set', get_Dataloader('dev',load='asr', use_gpu=self.paras.gpu, **self.config['solver']))
# Get 1 example for auto constructing model
for self.sample_x, _ in getattr(self,'train_set'): break
if len(self.sample_x.shape) == 4: self.sample_x = self.sample_x[0]
def load_data(self):
self.verbose('Loading testing data '+str(self.config['solver']['test_set'])\
+' from '+self.config['solver']['data_path'])
setattr(
self, 'test_set',
get_Dataloader('test',
load='asr',
use_gpu=self.paras.gpu,
**self.config['solver']))
setattr(
self, 'dev_set',
get_Dataloader('dev',
load='asr',
use_gpu=self.paras.gpu,
**self.config['solver']))
def get_dataloader(args, config):
if not os.path.exists(config['dataloader']['data_path']):
raise RuntimeError('[run_upstream] - Data path not valid:', config['dataloader']['data_path'])
print('[run_upstream] - Loading input data: ' + str(config['dataloader']['train_set']) + ' from ' + config['dataloader']['data_path'])
print('[run_upstream] - getting train dataloader...')
# select mode
try:
if config['transformer']['dual_transformer'] and config['transformer']['wave_transformer']:
raise ValueError('`dual_transformer` and `wave_transformer` can not both be True!')
except: pass
if 'dual_transformer' in config['transformer']:
load = 'dual_acoustic' if config['transformer']['dual_transformer'] else 'acoustic'
if 'wave_transformer' in config['transformer']:
load = 'wave_acoustic' if config['transformer']['wave_transformer'] else 'acoustic'
else:
load = 'duo' if bool(config['runner']['duo_feature']) else 'kaldi' if args.kaldi_data else 'acoustic'
# print path info
if load == 'duo':
print('[run_upstream] - Loading duo data: ' + str(config['dataloader']['train_set']) + ' from ' + config['dataloader']['target_path'])
elif load == 'kaldi':
print('[run_upstream] - Loading Kaldi data: ' + str(config['dataloader']['data_path']) + ' from these sets ' + str(config['dataloader']['train_set']))
elif load == 'wave_acoustic':
print('[run_upstream] - Loading wave data: ' + str(config['online']['libri_root']) + ' from these sets ' + str(config['dataloader']['train_set']))
elif load == 'acoustic' and 'online' in config:
print('[run_upstream] - Using online data from root: ' + str(config['online']['libri_root']))
elif load == 'acoustic':
print('[run_upstream] - Loading data: ' + str(config['dataloader']['data_path']) + ' from these sets ' + str(config['dataloader']['train_set']))
dataloader = get_Dataloader(split='train', load=load, use_gpu=args.gpu,
run_mam=True, mam_config=config['transformer'], **config['dataloader'], **config)
return dataloader
def load_data(self):
''' Load training / dev set'''
self.verbose('Loading text data from ' +
self.config['solver']['data_path'])
setattr(
self, 'train_set',
get_Dataloader('train',
load='text',
use_gpu=self.paras.gpu,
**self.config['solver']))
setattr(
self, 'dev_set',
get_Dataloader('dev',
load='text',
use_gpu=self.paras.gpu,
**self.config['solver']))
def load_data(self, split='train'):
''' Load data for training / testing'''
if split == 'train':
self.verbose('Loading source data from ' +
str(self.config.train_set) + ' from ' +
self.config.data_path)
elif split == 'test':
self.verbose('Loading testing data ' + str(self.config.test_set) +
' from ' + self.config.data_path)
else:
raise NotImplementedError('Invalid `split` argument!')
setattr(
self, 'dataloader',
get_Dataloader(split,
load='spec',
data_path=self.config.data_path,
batch_size=self.config.batch_size,
max_timestep=3000,
max_label_len=400,
use_gpu=True,
n_jobs=self.config.load_data_workers,
train_set=self.config.train_set,
dev_set=self.config.dev_set,
test_set=self.config.test_set,
dev_batch_size=1))
def load_data(self, split='train', load='phone'):
''' Load date for training / testing'''
assert(load in ['phone', 'cpc_phone', 'sentiment', 'speaker', 'speaker_large']), 'Unsupported dataloader!'
if load == 'phone' or load == 'cpc_phone' or load == 'speaker_large':
if split == 'train':
self.verbose('Loading source data from ' + str(self.config['dataloader']['train_set']) + ' from ' + self.config['dataloader']['data_path'])
if load == 'phone' or load == 'cpc_phone': self.verbose('Loading phone data from ' + str(self.config['dataloader']['train_set']) + ' from ' + self.config['dataloader']['phone_path'])
elif split == 'test':
if load != 'cpc_phone': self.verbose('Loading testing data ' + str(self.config['dataloader']['test_set']) + ' from ' + self.config['dataloader']['data_path'])
if load == 'phone': self.verbose('Loading label data ' + str(self.config['dataloader']['test_set']) + ' from ' + self.config['dataloader']['phone_path'])
elif load == 'cpc_phone': self.verbose('Loading label data from ' + self.config['dataloader']['phone_path'])
else:
raise NotImplementedError('Invalid `split` argument!')
elif load == 'speaker':
if split == 'train':
self.verbose('Loading source data from ' + str(self.config['dataloader']['train_set']).replace('360', '100') + ' from ' + self.config['dataloader']['data_path'])
elif split == 'test':
self.verbose('Loading testing data ' + str(self.config['dataloader']['test_set']).replace('360', '100') + ' from ' + self.config['dataloader']['data_path'])
else:
raise NotImplementedError('Invalid `split` argument!')
elif load == 'sentiment':
target = self.config['dataloader']['sentiment_config']['dataset']
sentiment_path = self.config['dataloader']['sentiment_config'][target]['path']
self.verbose(f'Loading {split} data from {sentiment_path}')
else:
raise NotImplementedError('Unsupported downstream tasks.')
setattr(self, 'dataloader', get_Dataloader(split, load=load, use_gpu=self.paras.gpu, \
run_mockingjay=self.run_mockingjay, mock_config=self.config['mockingjay'], \
**self.config['dataloader']))
def get_dataloader(args, config):
if not os.path.exists(config['dataloader']['data_path']):
raise RuntimeError('[run_upstream] - Data path not valid:',
config['dataloader']['data_path'])
print('[run_upstream] - Loading input data: ' +
str(config['dataloader']['train_set']) + ' from ' +
config['dataloader']['data_path'])
print('[run_upstream] - getting train dataloader...')
load = 'duo' if bool(config['runner']['duo_feature']
) else 'kaldi' if args.kaldi_data else 'acoustic'
if load == 'duo':
print('[run_upstream] - Loading duo data: ' +
str(config['dataloader']['train_set']) + ' from ' +
config['dataloader']['target_path'])
if load == 'kaldi':
print('[run_upstream] - Loading Kaldi data: ' +
str(config['dataloader']['data_path']) + ' from these sets ' +
str(config['dataloader']['train_set']))
dataloader = get_Dataloader(split='train',
load=load,
use_gpu=args.gpu,
run_mam=True,
mam_config=config['transformer'],
**config['dataloader'])
return dataloader
def load_data(self, split='train'):
''' Load data for training / testing'''
if split == 'train':
self.verbose('Loading source data ' +
str(self.config['dataloader']['train_set']) +
' from ' + self.config['dataloader']['data_path'])
if self.duo_feature:
self.verbose('Loading target data ' +
str(self.config['dataloader']['train_set']) +
' from ' +
self.config['dataloader']['target_path'])
elif split == 'test':
self.verbose('Loading testing data ' +
str(self.config['dataloader']['test_set']) +
' from ' + self.config['dataloader']['data_path'])
else:
raise NotImplementedError('Invalid `split` argument!')
if self.duo_feature:
setattr(self, 'dataloader', get_Dataloader(split, load='duo', use_gpu=self.paras.gpu, \
mam_config=self.transformer_config, **self.config['dataloader'])) # run_mam is automatically performed
else:
setattr(self, 'dataloader', get_Dataloader(split, load='acoustic', use_gpu=self.paras.gpu, run_mam=True, \
mam_config=self.transformer_config, **self.config['dataloader']))
def load_text(self, data_config):
# Independent training set for CLM
self.train_set = get_Dataloader('text', text_only=True, **data_config)
self.data_iter = iter(self.train_set)
