python GMMHMM clarification

Question

I am reading about gaussian mixture hmms and I want to implement one. I am using the hmmlearn package and I know it has been deprecated from scikit-learn. In the documentation it says that in the GMMHMM constructor it requires: "gmms (array of GMM objects, length n_components) GMM emission distributions for each state."

The way I understand I should proceed is the following : 1) First perform the GMM training by using the scikit-learn 2) Pass the GMM objects to the GMMHMM constructor in order to train the hmm.

Is this correct or am I missing something?

Thanks.

Answer 1

No, GMMHMM will fit the mixtures automatically. All you need to do is to specify the desired number of states in an HMM and the number of components in each mixture.