NAN problem in HMM

Hi,

I have 40 states in the transition matrix and 40 emissions which represents 1 for each state.

When it comes to insert sequences to get the log of the probability of sequence SEQ given the transition matrix, TR and the emission matrix, E (using HMMDecode , Baum Welch Algorithm), 4224 out of 31000 sequences are given NAN. Is there any way to solve this?

I am thinking to add noise to transition matrix since there a lot of states which has 0% probability to go to another state.