An inference algorithm for hidden Markov models which computes the posterior marginals of all hidden state variables given a sequence of observations/emissions. It uses dynamic programming to compute the values in two passes, the first going forward in time and the second backward.