Research On Cerebellum Model And Its Contributions To Speech Acquisition And Production Based On DIVA Model

In order to explore the processes behind speech acquisition and production, it is crucial todescribe and achieve a functional representation of areas in the brain that are involved in speechproduction and speech perception in neuropsychology and neuroanatomy level. And this researcharea has become a hotspot. DIVA (Directions Into Velocities of Articulators) model is an adaptiveneural network model and the most comprehensive current model of speech acquisition andproduction which has biophysical significance. The model is a mathematical model that learns tocontrol movements of a simulated vocal tract in order to produce words, syllables, or phonemes.Otherwise, experimental observations point to the involvement of the cerebellum in a wide varietyof motor and cognitive tasks. The cerebellum has a uniquely uniform neurophysiologicalarchitecture and widespread connectivity to and from cerebral cortex. Therefore, the intent of theresearches that need to do on DIVA model at present is to explore the role of the cerebellum incoordinating the temporal characteristics and other aspects during speech production, in addition tofocusing on the neural control of the production of elementary phonetic units.This thesis firstly introduces the DIVA model, including its basic principles, compositions, andlearning process, then explores cerebellar contributions to DIVA model for different aspects ofspeech acquisition and production. On this basis, the thesis extends the DIVA model, adding thecorresponding cerebellum modules to the projections from speech sound map to auditory andsomatosensory target maps and the projection from feedback control map to articulator velocity andposition maps in the feedback control subsystem. For computational and implementation simplicity,a “functionally equivalent” cerebellum was embedded, not a complete neurophysiologicallyplausible model of the cerebellum. The experimental results indicate that the extended DIVA modelcan produce more clear and explicit speech sounds, and the formant frequency curves of speechsignal are smoother.The thesis also constructs a cerebellar neuronal network model for DIVA model combined withthe cerebellum anatomical structure and knowledge of neurophysiology, and then applies to thefeedforword control subsystem of DIVA model. The results show that the new cerebellum modelcan solve the problems of the learning and maintenance of feedforword motor commands.