A Computational Model on the Emergence of Vowel Transition Perception

Abstract

The speech generalizing abilities of infants in their native language improve as they grow up throughout their first half-year [1]. The emergence of this ability is a self-organizing process because the access of word meaning of the young infants is still very limited. Guenther and Gjaja simulated this phenomenon successfully by using self-organizing maps [2]. In their models, however, only static features are considered as inputs, and internal temporal relationships within speech sounds are not considered. In the present study, we simulate the same phenomenon by modifying a word recognition model that takes temporal features into account [3,4]. A new self-organizing algorithm, relying on the competition among dendrites or neurons, is used in our model.