An isolated word recognizer system based on corrective training

A corrective training method of the gradient type which is based on the modification of the state transition probabilities is developed. To increase the discrimination between two HMMs (hidden Markov models) λ₁ and λ₂, Viterbi´s algorithm is used to segment the sequence of observations, obtaining for the state i and the sequences O⁽¹⁾ and O⁽²⁾ the permanencies in the state i: n_i⁽¹⁾ n_i⁽²⁾, respectively. With this value, the statistics `of the model are estimated. After a few iterations an acceptable convergence is obtained