Created by: xinghai-sun

resolve #113 (closed)

1. Improve audio featurizer (resample, db_normalize, and random shift), as suggested in the speech_dl codes.
2. Add shift augmentor.
3. Update default arguments to be the current best seggestion.
4. Add checkpoints with pass id.

Training experiment is in progress, and its results will be pasted here ASAP.