Initializing HMM parameters

Next, initialize a set of HMM parameters using the commands.
   names={'MEAN','STDV'};
   min_std=[.1 .1];
   NSTATES=3;
   NMODE=10;
   parm=init_hmm(x,NSTATES,NMODE,names,min_std);
This first two commands define the feature names and the minimum standard deviations for Gaussian mixture estimation (See Section 13.2). The initial HMM parameters are obtained by using software/init_hmm.m which creates a uniform state transition matrix $ A$ and prior probability $ \pi$. The PDF of the feature vector in each state is approximated by Gaussian mixtures. The starting point for the Gaussian mixture parameters are obtained by the function software/init_gmix.m described in the previous sections.



Baggenstoss 2017-05-19