140616 - Xiaoxi Wang

Last week:

trained LMs and test with new AM and test set

This week:

build LMs and interpolate with human voice LM and out-domain LM. Then test WER, compare with results from SinoVoice