How to infer with the custom audio?

Hi, thanks for your work. I run demo successfully, but it seems to be limited into given dataset. 
I'm trying to infer a video result with my custom audio. Could you give some guidance?