The Study Of Vocal Tract Model And Its Control Mechanism Based On The Spech Production And Acquisition Of DIVA Model

With the development of artificial intelligence, the research on this field has processed a lot.Control on the speech production and acquisition like human is a key problem of the robot soundsystem urgently to be solved. DIVA ( Directions Into Velocities of Articulators ) is a mathematicalmodel of the processes on speech acquisition and production, it is also an adaptive neural networkthat learns to control movements of a simulated vocal tract in order to produce words, syllables orphonemes. In today's real biologically meaningful speech acquisition and production neuralnetwork model, the DIVA model is the most thorough and the only one that applies pseudo-inversecontrol scheme.This paper first states the development of DIVA model and vocal tract model , then introducesthe theory of DIVA model, including the concept and basic structure of DIVA model , studies anddiscusses the map between Orosensory Direction Vector and Articulator Velocity Vector and themechanism of pseudo-inverse control scheme. This paper proposes a redistribution advancedmethod based on zero space to improve the pseudo-inverse control scheme of DIVA model. Thismethod avoids articulator entering into saturation statement too early, and improves the robustnessof DIVA model.The paper imports EMA (electromagnetic articulograph) database to describe the Maedageometry vocal tract model. The author uses an adaptive process to describe the method of applyingEMA data to DIVA model, which can be more precisely estimate the shape of vocal tract. Based on/u/ phoneme, in the condition of perturbations of the upper jaw, the author analyzes the relationshipbetween lip protrusion and constriction size ,discusses motor equivalent and the configuration ofvocal tract parameters of DIVA model. Finally, the author summarizes the research contents, pointesout the direction of subsequent research, and raises a new thinking for further study.