Using sliding windows to fit the HMM can lead to very long (in time) and heavy (in cpu/memory) processes. One simple way to speed up t process, is to split the inference by chromosome.