Research On Applying AGNN To Classify EEG Based On The DIVA Model

Abstract: In recent years, the main ideas of the artificial speech synthesis system to pursuit isthat description and simulation of the human brain function involved in speech understandingand generation region in neuropsychology and neuroanatomy level. Around this main idea, manyresearchers make a huge effort in the exploration and research of speech acquisition andgeneration computing model, and gain fruitful results. One of the mosts representative andgroundbreaking is DIVA (Directions Into Velocities of Articulators) model which is researchedby professor Gunther (Frank H. Guenther) of speech lab at Boston University and his team. It isa neural network model, it can describe related process of speech acquisition and generation,generate words, syllables or phonemes by controlling an analog channel besides. Testing anddefinition of the DIVA model is the most thorough in today’s speech acquisition and generationmodel by comparing with other models. It has biological significance and is a kind of model thatapply pseudo-inverse control scheme.First, the paper introduce the development process, the present situation of the DIVA modeland the neural network classifier. Then the related theory of DIVA model is classified in detailincluding the basic concept, structure and the learning stage. Upon above study, the detailedwork process of each component in learning stage and speech process is proposed through thesimulation of the DIVA model. A specific word is used as an example to realize the specificlearning process of that word.Second, by using the international “Brain Computer Interface”(BCI) competitionexperimental methods and data, according to the characteristics of EEG(electroencephalogram)signals, the article compared the pros and cons of both Fisher linear discriminant andAGNN(Adaptive Growth Neural Network) classification through simulation experiments basedon describing the model’s neural network classifier.Finally, the paper gave specific execute solution that applying Adaptive Growth NeuralNetwork to the DIVA model, and through simulation results compared the efficiency of RBF andAGNN using to the DIVA model which is realized by the simulation program. At the end of thearticle, the author summarized the content of the research and pointed out the direction of futureresearch, pioneered the idea of further research.Studies done in the paper can make a significant contribution to improve the validity and reliability of the DIVA model, promote the development of neuropsychology and neuroanatomyand accelerate research on medical devices.