论文部分内容阅读
将扩展联想记忆神经网络用于噪音语音识别，它是在一多层感知器（ＭＬＰ）后接一反馈网络构成。对于反馈联想网络，我们提出了基于误差反传的快速梯度学习算法，从而提高了网络的联想记忆能力和训练速度。通过改变误差能量函数，使权值修正根据误差的大小而改变，并导出相应的快速算法，使扩展联想网络的训练速度提高３～５倍，整个系统具有很高的自适应性、鲁棒性、容错性和联想记忆能力。最后根据语音识别结果提取概念，进行自组织拓扑语义影射，在平面上反映出词汇间的语义关系。
The extended associative memory neural network is used for noise speech recognition. It is formed by a multi-layer sensor (MLP) followed by a feedback network. For the feedback association network, we propose a fast gradient learning algorithm based on error feedback, which improves associative memory and training speed of the network. By changing the error energy function, the weight correction is changed according to the size of the error, and the corresponding fast algorithm is derived to improve the training speed of the extended associative network by 3 to 5 times. The whole system has high self-adaptability and robustness , Fault tolerance and associative memory ability. Finally, according to the result of speech recognition, the concept is extracted and the self-organizing topological semantic mapping is performed to reflect the semantic relationship between words in the plane.