Optimizing the likelihood function of the HMM is analytically
intractable, but the iterative Baum-Welch algorithm [8] is a
conventional efficient method to estimate the HMM parameters
and one of the most important reasons for the popularity of ML
training in HMMs.