Most of the time, you probably won't have a good alignment already, or you might have an alignment for only a few sequences but not all of them. In this case, you would run hmmt (``HMM train'') and train your HMM on a number of unaligned example sequences.
Training an HMM is an iterative process that seeks to maximize the probability that the HMM accounts for the example sequences. There is a ``local maximum'' problem; the model, though converged to some locally optimal choice of parameters, is not guaranteed to be the best possible model. In an attempt to avoid this, hmmt uses a simulated annealing procedure by default.