The Baum-Welch algorithm[13] is used to estimate parameters of hidden Markov models from data. It is based on forward and backward procedures. These can be adapted for LOHMMs with few differences. The Baum-Welch algorithm is an instance of the expectation-maximisation (EM) algorithm. In the E-step, the expected values of the latent variables are estimated keeping the parameters constant, and in the M-step, the parameters are updated keeping the latent variables constant. The two steps are iterated until a convergence criterion is fulfilled. It has been shown [2] that the EM-algorithm converges to a fixed point. Let us now discuss these two steps.