YangKang
There is a bug in line 33,and occurs TypeError: 'int' object is not callable.Because when run in python,the "voice_index" is identified as "int" object instead of "voice_index" method. 
And you should modify line 26 in "\_\_main\_\_.py"，the last args should be "args["language"]" 
It should be chinese instead of mandarin  Beacuse in your code is "chinese"  so there should be "chinese". 
When using the Chinese song ，the performance is worse.Maybe the file"lyrics_tokenizer.py" caused the results. Because chinese words doesn't need to tokenize into subwords
I fix this problem，because the musescore decode the midi into .xml is wrong. Thanks for your help and project!
> Python3.6 + pyltp 0.1.9.1/0.2.1 + ltp model 3.3.1/3.4.0无法复现README的样例 Python2.7 + pyltp 0.1.9.1/0.2.1 + ltp model 3.3.1也无法复现README的样例 有没有人能复现README的样例的？ 可以用ltp，然后改一下代码可以复现
Maybe I find the reason that the size of unlabeled data we downloaded from "ftp.ncbi.nlm.nih.gov/pubmed/baseline" is 19G.According to your paper,should I sample the unlabeled data and then convert it to...
> It has to be in its own block, i.e. "display mode" of KaTeX. The `equation` environment can't be rendered inline. I check the code, and found that the inline...
> I also had this problem when using \tag in formulas. The \tag only works when 'displayMode' is true. And I fixed this problem and PR. Could you please tell...