def main(args):
deepspeech = setup_deepspeech(CONFIG_PATH,
ALPHABET_PATH,
pretrained_weights=args.pretrained_weights)
train_generator, dev_generator = create_generators(deepspeech, args)
deepspeech.fit(train_generator,
dev_generator,
epochs=args.epochs,
shuffle=False)
deepspeech.save(WEIGHTS_PATH)
if __name__ == "__main__":
chdir(to='ROOT')
parser = create_parser()
ARGUMENTS = parser.parse_args()
LOG_PATH = os.path.join(ARGUMENTS.model_dir, 'training.log')
CONFIG_PATH = os.path.join(ARGUMENTS.model_dir, 'configuration.yaml')
ALPHABET_PATH = os.path.join(ARGUMENTS.model_dir, 'alphabet.txt')
WEIGHTS_PATH = os.path.join(ARGUMENTS.model_dir, 'weights.hdf5')
logger = create_logger(LOG_PATH,
level=ARGUMENTS.log_level,
name='deepspeech')
logger.info(f'Arguments: \n{ARGUMENTS}')
main(ARGUMENTS)
def run_consumer(queue):
""" Run program in the infinite loop (a daemon process). """
while True:
try:
command = next_in(queue)
if command:
logger.info(f'Program execute the command: {command}')
execute(command)
logger.info(f'Correct run: {command}')
else:
time.sleep(10)
except subprocess.CalledProcessError:
logger.error(f'Program can not execute the command')
if __name__ == "__main__":
chdir(to='ROOT')
parser = argparse.ArgumentParser()
parser.add_argument('--queue', required=True, help='The queue with tasks to execute')
parser.add_argument('--python', required=True, help='The interpreter path')
args = parser.parse_args()
logger = create_logger(name='consumer')
logger.info('The consumer has been started')
logger.info(f'Execute the queue: {args.queue}')
run_consumer(queue=args.queue)
logger.info(f'Model contains: {units//1e6:.0f}M units ({units})')
metrics = evaluate(deepspeech, generator, save_activations, store_path)
logger.info(f'Mean CER: {metrics.cer.mean():.4f}')
logger.info(f'Mean WER: {metrics.wer.mean():.4f}')
if __name__ == '__main__':
ARGUMENTS = parse_arguments()
chdir(to='ROOT')
CONFIG_PATH = os.path.join(ARGUMENTS.model_dir, 'configuration.yaml')
ALPHABET_PATH = os.path.join(ARGUMENTS.model_dir, 'alphabet.txt')
WEIGHTS_PATH = os.path.join(ARGUMENTS.model_dir, 'weights.hdf5')
logger = create_logger(ARGUMENTS.log_file,
level=ARGUMENTS.log_level,
name='evaluate')
logger.info(f'Arguments: \n{ARGUMENTS}')
main(store_path=ARGUMENTS.store,
features_store_path=ARGUMENTS.features_store,
batch_size=ARGUMENTS.batch_size,
save_activations=ARGUMENTS.save_activations,
mask=ARGUMENTS.mask,
mask_F=ARGUMENTS.mask_F,
mask_mf=ARGUMENTS.mask_mf,
mask_T=ARGUMENTS.mask_T,
mask_mt=ARGUMENTS.mask_mt,
mask_ratio_t=ARGUMENTS.mask_ratio_t)
logger.info(f'Program execute the command: {command}')
execute(command)
logger.info(f'Correct run: {command}')
else:
time.sleep(10)
except subprocess.CalledProcessError:
logger.error(f'Program can not execute the command')
if __name__ == "__main__":
chdir(to='ROOT')
parser = argparse.ArgumentParser()
parser.add_argument('--queue',
required=True,
help='The queue with tasks to execute')
parser.add_argument('--python', required=True, help='The interpreter path')
parser.add_argument('--log_file', required=True, help='The log file')
parser.add_argument('--log_level',
type=int,
default=20,
help='The log level (default set to INFO)')
args = parser.parse_args()
logger = create_logger(args.log_file, args.log_level, name='consumer')
logger.info('The consumer has been started')
logger.info(f'Execute the queue: {args.queue}')
run_consumer(queue=args.queue)
parser.add_argument('--store',
required=True,
help='File hdf5 keeps evaluation results')
parser.add_argument('--model_dir',
required=True,
help='Pretrained model directory')
parser.add_argument('--features_store',
required=True,
help='HDF5 Store with precomputed features')
parser.add_argument(
'--batch_size',
type=int,
required=True,
help=
'Batch size (depends on model) should be the same as during training')
parser.add_argument('--log_file', help='Log file')
parser.add_argument('--log_level', type=int, default=20, help='Log level')
parser.add_argument('--save_activations',
type=bool,
default=True,
help='Save all activation through evaluation')
args = parser.parse_args()
chdir(to='ROOT')
logger = create_logger(args.log_file,
level=args.log_level,
name='evaluate')
logger.info(f'Arguments: \n{args}')
main(args.store, args.model_dir, args.features_store, args.batch_size,
args.save_activations)
deepspeech.load(args.pretrained_weights)
train_generator = deepspeech.create_generator(
args.train,
batch_size=args.batch_size,
source=args.source,
shuffle_after_epoch=args.shuffle_after_epoch)
dev_generator = deepspeech.create_generator(args.dev,
batch_size=args.batch_size,
source=args.source)
deepspeech.fit(train_generator,
dev_generator,
epochs=args.epochs,
shuffle=False)
deepspeech.save(weights_path)
if __name__ == "__main__":
chdir(to='ROOT')
arguments = parse_arguments()
log_path = os.path.join(arguments.model_dir, 'training.log')
config_path = os.path.join(arguments.model_dir, 'configuration.yaml')
alphabet_path = os.path.join(arguments.model_dir, 'alphabet.txt')
weights_path = os.path.join(arguments.model_dir, 'weights.hdf5')
logger = create_logger(log_path,
level=arguments.log_level,
name='deepspeech')
logger.info(f'Arguments: \n{arguments}')
main(arguments)