Research On The Chinese Speech Pronunciation System Of DIVA Model

In recent years, it is the key idea of designing the speech synthesis system to describe and simulate the functions corresponding to the associated brain areas involved in speech production and acquisition at the level of neuroanatomy and neuropsycology. Around this topic,great efforts were made in researching and discussing the computational model of speech production and acquitision.The most outstanding model is the DIVA(Directions Into Velocities of Articulators) model which is proposed by the professor Frank Guenther and his colleges in speech lab, Boston University.The DIVA model is a neural network model describing the process of speech production and acquisition to product word, syllable, and phonemes by controlling a simulated vocal tract. However, DIVA model takes 29 underlying phonemes in English as research background, which makes the model not adapt to Chinese voice generation and acquisition.First, to solve the problem that the speech region of interest(ROI) is based on the English language research,this article conduct functional magnetic resonance imaging experimental study of Chinese speech processing, and analyse the experimental results. We found that the right temporal lobe and the adjacent edge of the rear upper back, bilateral ventral occipital region appeared different activation states, compared with the English pronunciation research.Second, to solve the problem that EEG (electroencephalogram) data collected through non-invasive Brain-Computer Interface shows low resolution and large interference in the research of Chinese Neuralynx System, this article proposes a constraint processing method for the EEG signal based on DIVA model.Experimental results show that on condition of spatial information of points which are activated by fMRI data, it locates ROI precisely.This article provides a feasible solution for electroencephalogram signal problem handling in Chinese Neuralynx system and lays a foundation on related studies of future Chinese voice generation and acquisition.