Found the problem, seems setting the cycles to 10000 fixed the problem. I had kept them on 20000 the whole. Not entirely sure how the cycles could affect soundcard detection but I guess they can.

A lot of poorly designed games will try to detect the sound card by sending some data to it and then waiting for a response, and on faster computers (or emulators running at higher cycle counts) a game will sometimes give up waiting before the sound card gets around to sending data back. I remember that some Sierra games had that issue at one point, and that updated Sierra driver files were released to fix it.

I wonder if a function could be included in DosBox that would test all emulated sound cards. It would automatically cycle DosBox to determine the # of cycles required for sound card detection.

It's a good thought, but there are probably almost as many detection methods used by games as there are games out there... It would be interesting (and useful for multiple situations) for someone to write a program that tries to detect cards by multiple methods to see if some are failing and others not.