语音处理的中心激励侧向抑制网络模型  

On-center Off-surround Network Model on Speech Processing

作　　者：李锋[1] 韦君银[1] 

机构地区：[1]东华大学计算机学院,上海200052

出　　处：《计算机工程》2008年第1期1-3,共3页Computer Engineering

基　　金：上海市科委基金资助项目(06bz15005)

摘　　要：根据人类听觉神经系统的结构,提出了一个"中心激励、侧向抑制"的网络模型。在该模型中,输入层的每一个神经元细胞在激励输出层对应的细胞的同时,会对其输出层其他邻近细胞起到一定的抑制作用。另外输出层的每个神经元也具有负反馈功能。该文给出了该网络模型的数学方程和网络参数调整的学习算法,证明了该算法的收敛性。该网络模型在声音处理上保留了人耳的优势,可以自动过滤噪声,搜索声音的中心频率段,将声音的物理频谱转换成相应的生理反应模型。通过仿真数据解释了人耳特有的频道抑制现象。On the base of auditory neural system, the on-center off-surround network model on processing of the sound wave is presented. In the network model, the neuron in the input layer excites the corresponding neuron in the output layer meanwhile it inhibits the neighborhood neuron. In the output layer, the neuron production auto-attenuates at a definite speed too. The mathematic equation and the learning algorithm on the neural network are also presented. The learning algorithm convergence is proved. The network has its advantage in sound wave processing. In addition, it can filter the noise, search the significance frequency segments and transform the speech physical spectrum to the biophysics action model. The ＂channel suppresser＂, the human acoustical system special phenomenon, is explained based on the model.

关 键 词：神经网络 频道抑制 噪声过滤 语音识别 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]