I have been reading some literature about HMM and for what I know the Baum-Welch algorithm can be used for training a HMM model. So my question would be, which are the minimum components that I need to have for training a HMM model?
In my case I know the hidden states, the observation states; and the emission probabilities which I obtained from counting the frequencies of occurrences of a particular event. What can I do if I do not have the transition probabilities and the initial probabilities for the starting states? Does the BM algorithm can work only with the states and probabilities observations for inferring the information missing?