[waiquoo]

Winston Timp has done some interesting work in this area (http://www.sciencedirect.com/science/article/pii/S0006349512...). Basically, by training a hidden Markov Model, you can get the most likely sequence from a noisy source.