2

Reason:
1. Somebody will rewrite text from paper and can't watch on screen and paper in the same time, so he/she will look on paper, not on screen. Sound on misspelled word will allow to eliminate bug in written text.
2. Not each person can write text without watching on keyboard, so on misspelled word sound will be played.

Of course - word will be checked on white space insertion, like space, tab, enter, etc.

I don't know how much this will get used over just marking the misspelled words and making sure you go back and review that after you've written what you've written. For the user who is copying text from a paper, ignoring mistakes until the end will be quicker because they can establish a good pace by just focusing on typing what's on the page.
I could see this possibly being useful for some of the users who keep their eyes on the keyboard because they haven't learned to touch-type. (though they should still look up and double-check what they've written every once-in-a-while)