Estimation method of hidden Markov model , estimation apparatus and the estimated program .

摘要

Calculation of unknown parameters to obtain a possible hidden Markov model estimation algorithm even conditions other than special. As unknown parameters of the hidden Markov model, estimating means 14, the likelihood P state transition probability a, the output probability b, and initial state probability π | and each output probability amount of (y θ), and the expected value of N _a state transition and update setting means 24 to update setting the expected value N _b of, each probability amount of just before you update set in the update configuration means 24 and not only the expected value, each probability the amount of time-shifted earlier and each expected value as well used, using the observed data read from the register 1, and the value β of the acceleration parameters set in the initial setting unit 22, each new probability weights and the respective expected values ​​by applying a small approximation by Taylor expansion I and a computing means (26) for calculation.