Simulation Of Neural Mechanism For Speech Perception With DIVA Neural Network Model

The results of the psycholinguistic experiments shows that the perceptual magneteffect is the important factor in infants’ speech development. This effect produces awarped auditory perceptual space, resulting the sound around one phoneme will beclassified as the same sort. The purpose of this study is to develop a neural networkmodel in simulation of speech perception and speech acquisition. The neural networkmodel with unsupervised learning is used to determine the phonetic categories ofphoneme according to the formant frequencies of the vowels, therefore it can simulatethe process of speech acquisition from auditory.The modified “Self-Organizing Map”(SOM) algorithm is proposed to producethe auditory perceptual space of English vowels and are compared with findings frompsycholinguistic experiments. Simulated results show model makes the categorizationof English/r/and/l/and prototype and non-prototype vowels and forms auditoryperceptual space.In addition, this speech perception model was combined with the neural networkmodel (DIVA model) to simulate categorization of ten English vowels to show thelearning capability of speech perception and production. We further extended thismodified DIVA model to show its capability to categorize Chinese vowels and theirproductions. Finally, this study proposed further development and related discussionsfor this speech perception model and its clinical application.