I don't think the QST2 method is your best chance to get the transition state, as it will require having the acetaldehyde-ammonia pre-reaction complex in a proper orientation so it can be properly interploated.
I recommend you to optimize the hemiaminal product, in a conformation that places the O-H bond pointing towards the nitrogen, and then perform a relaxed scan elongating the C-N distance. You should get a maximum that corresponds to the dissociation / bond formation, take that structure as your initial guess for optimizing the transition state.