In continuous speech, the feature of phoneme is ambiguous because of the coarticulation effect, which causes a difficulty of an automatic recognition of continuous speech. In the other hand, there is no coarticulation problem in human auditory.In order to make the recognition easier, we propose an associative model of the human auditory process. It is assumed that the observed phoneme is associated with the acoustic characteristics of the three points (preceding, observed, and following points), and the association is linear.The model was tested and it was found to be effective in a high noise condition. It was also found that the elements of the memory matrix were formed to clear the distinction between similar vowels, and that the spectrum reproduction error was sufficiently small.
展开▼
