- Change
data_root
inconfigs/config.json
file - In
train.py
, we need to change 2 lines:- config.vocab_type = 'xxx' (xxx in ['ctc', 'attention', 'joint'])
- model = XxxModel(config)
- Please take a look at
data/val.json
file, the data in this file is a dictionary with following structure:
{
path1: label1,
path2: label2,
...
pathn: labeln
}
- Run
run_sagemaker.ipynb
file with jupyter notebook on SageMaker.