Created by: lixx0105

hi ,

The following python version ctc beam search result is not correct for mandarin with and without LM scorer.(https://github.com/PaddlePaddle/DeepSpeech/blob/develop/decoders/decoders_deprecated.py)

I found this code is for English word based LM which can explain the result with LM is not correct for mandarin. But result without LM is not correct either. C++ version is correct.

Can you give some idea why the python version is not correct. thanks in advance.