摘要
脑电信号获取过程中,工频噪声干扰现象往往会使所获取的信息产生多种多形态瞬时结构波形,这种现象影响到DIVA(Directions Into Velocities of Articulators)模型对语音的正常处理.为此,本文提出了一种面向特征提取的脑电信号结构自适应稀疏分解模型,并在此基础上,通过采用匹配追踪算法求解最佳原子、使用过完备原子库中原子表示原始脑电信号等方法,实现了信号去噪的目的,效果好于传统的小波变换去噪方法.仿真实验表明,本文提出的方法提高了DIVA模型语音发音的精度.
There are power frequency interference and other kinds of noise in the electro encephalo gram (EEG) signal acquisition process. They make the signal show non-stationary and a variety of multi-form waveform in the instantaneous structure. Then such signal will affect the normal processing of the speech in DIVA(Directions Into Velocities of Articulators) model. Therefore, this paper proposes an adaptive sparse decomposition model for the feature extraction of EEG signal structure and makes use of Matching Pursuit algorithm to solve the optimal atom. Then the original EEG signal can be represented by atoms in the complete atomic library. Fmally, this model removes noise that exists in the EEG signal and is compared with wavelet transform method. Simulation results show that after we put the denoising EEG signal into the model, the phonetic pronunciation improves.
出处
《电子学报》
EI
CAS
CSCD
北大核心
2015年第4期700-707,共8页
Acta Electronica Sinica
基金
国家自然科学基金(No.61373065
No.61271334)
关键词
DIVA模型
脑电信号
噪声
稀疏分解
directions into velocities of articulators (DIVA) model
electro encephalo gram (EEG) signal
noise
sparse decomposition