def main():
print_arguments(args)
print('开始生成数据列表...')
create_manifest(annotation_path=args.annotation_path,
manifest_path_prefix=args.manifest_prefix)
print('开始生成数据字典...')
counter = Counter()
count_manifest(counter, args.manifest_path)
count_sorted = sorted(counter.items(), key=lambda x: x[1], reverse=True)
with open(args.vocab_path, 'w', encoding='utf-8') as fout:
labels = ['?']
for char, count in count_sorted:
if count < args.count_threshold: break
labels.append(char)
fout.write(str(labels).replace("'", '"'))
print('数据字典生成完成！')
print('开始抽取%s条数据计算均值和标准值...' % args.num_samples)
compute_mean_std(args.manifest_path, args.num_samples, args.output_path)
def main():
print_arguments(args)
# 创建保存模型的文件夹
if not os.path.exists(args.save_model_path):
os.makedirs(args.save_model_path)
# 加载数据字典
with open(args.vocab_path, 'r', encoding='utf-8') as f:
vocabulary = eval(f.read())
vocabulary = "".join(vocabulary)
# 获取模型
model = GatedConv(vocabulary)
# 加载预训练模型
if args.restore_model:
model = torch.load(args.restore_model)
model = model.cuda()
train(model=model,
train_manifest_path=args.train_manifest_path,
dev_manifest_path=args.dev_manifest_path,
vocab_path=args.vocab_path,
epochs=args.epochs,
batch_size=args.batch_size,
learning_rate=args.learning_rate)
type=str,
help="language model path. (default: %(default)s)")
parser.add_argument("--dev_manifest_path",
default="dataset/manifest.dev",
type=str,
help="train manifest file path. (default: %(default)s)")
parser.add_argument("--vocab_path",
default="dataset/zh_vocab.json",
type=str,
help="vocab file path. (default: %(default)s)")
parser.add_argument("--batch_size",
default=64,
type=int,
help="number for batch size. (default: %(default)s)")
args = parser.parse_args()
print_arguments(args)
alpha = 0.8
beta = 0.3
cutoff_top_n = 40
cutoff_prob = 1.0
beam_width = 32
num_processes = 4
blank_index = 0
model = torch.load(args.model_path)
model = model.cuda()
model.eval()
# 创建解码器
decoder = CTCBeamDecoder(model.vocabulary, args.lm_path, alpha, beta,