Training a hidden Markov model through expectation-maximization, using Baum-Welch formulae, for applications in speech recognition.
A matrix, B matrix and time steps must be added to read_initialmatiricies before use.
Run in the following order:
- read_initialmatricies
- calc_B
- calc_forward
- calc_backward
- get_occupationlikelihoods
- reestimate_meanvariance
- gauss_graph
- get_transitionlikelihoods
- reestimate_A
File retrain.m may be used to perform many iterations of above procedure.