| With the deepened research on brain functional imaging, human has shared certain common view on the mechanism of speech motion control. On the basis of that, the research group led by the professor Guenther in Boston University put forward a neural computational model DIVA(Directions Into Velocities of Articulators). It can explain and describe the process about the speech production and acquisition. And the ElectroncephaloGram(EEG) signal is inputted into DIVA model.However, it has non-stationarity and instantaneous waveform structure of various forms and is affected by various noises easily in the gathering process. Then it influences speech processing ability of the DIVA model. Therefore, based on the thought of sparse decomposition, a construction method of the over-complete dictionary specifically for EEG signals structure is proposed instead of original Gabor dictionary in order to de-noising and then to improve the speech learning ability of DIVA model.Firstly, this paper introduces the basic principles of DIVA model and relationship between the various components. Then it discusses the basic features of EEG signal as the model inputting. Meanwhile, it instructs the principles and advantages and disadvantages of the traditional EEG de-noising methods specially.Then, constructed steps of the new sparse decomposition dictionary is described detailed. Simulation experiments show that the dictionary has better sparsity and reconstruction effect to EEG signal than the traditional Gabor dictionary. After the analyzing the de-noising principle about sparse decomposition, it applies the dictionary to eliminate the confounding noise in the EEG signal. Compared with the traditional wavelet de-noising method, de-noising effect is more obvious.At last, aiming at the deficiency existed in current Speech-Somatosensory Mapping, this paper uses mixed algorithm by combing self-organizing mapping and particle swarm optimization. Inputting de-nosied EEG signal to DIVA model improves effectively clustering effect and accuracy of model when learning Chinese vowel pronunciation.The research improves DIVA model performance, and lays a good foundation for its application on the Chinese pronunciation, and plays an important role in solving the language function barrier in the future. |
