PARALLEL PROCESSING METHOD FOR LEFT-TO-RIGHT TYPE HIDDEN MARKOV MODEL METHOD

摘要

PURPOSE: To improve the parallel processing efficiency of a left-to-right type hidden Markov model HMM method which is frequently applied to the recognition of voices and graphics and can attain the processing necessary for the real-time properties of the recognition of voices and graphics at a high speed in a parallel processing way. ;CONSTITUTION: An array processor transfers the data between the contiguous processing elements or to the outside and carries out the desired calculation after connecting the same processing elements in a ring shape. Then the learning processing and the searching processing are carried out in parallel to each other in accordance with the transition width with use of the array processor and a left-to-right type HMM method. In this case, the transfer direction of data transferred between the processing elements is selected by the type of a probability parameter that must be obtained in the learning/searching processing applying the HMM method. Then a 1st data train is transferred between the processing elements in the selected transfer direction and in the frequency accordant with the transition width. A 2nd data train is inputted from outside synchronously with the transfer of the 1st data train. Then the desired calculation processing is carried out with use of both 1st and 2nd data trains.;COPYRIGHT: (C)1993,JPO&Japio