If you think that the audio will only be linked to the entry in one level, then I think that’s not correct, and that the method I mentioned will work exactly the same when audio is attached. Or do you see a reason why it wouldn’t?
As I have written elsewhere: you should see a course as a database with many entries. For example:
ni hao - hello - <audio ‘ni hao’> - … - … (other details)
bu - not - - … - … (other details)
Once a level is created, in fact a list of pointers (as I call them) is compiled, each pointing to an entry. The audio is part of the entry. And so is your learning history/progress.
So when you add ‘ni hao’ to one level and to anther, both will have the same audio, as it is part of the same entry. Unless of course, you create a complete new entry - in that case you will see ‘hello’ twice in the database.
Does that indeed answer your concern or is you concern actually different?