Hidden Markov Model: Algorithms
Hidden Markov Model: Algorithms
- Training
- Expectation-Maximisation Algorithm (Baum-Welch)
- Estimate expected transitions based on training data and HMM model l = (A,B,p)
- Re-estimate new model l from the expected transitions of current model and training data
- Testing
- Dynamic Programming Method (Viterbi Algorithm)
- Calculate P(O|l) for each model l and the testing data utterance O, by finding the best state sequence
- Maximum-Likelihood (ML) estimate of unknown testing data is to pick the model l which yields the maximum P(O|l)
- Reference:
- L.R. Rabiner, A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition, Proc. IEEE, Feb 1989, Vol. 77, No. 2, pg 257-286