基于SOFM神经网络的IP电话语音压缩编码设计

对自组织特征映射（SOFM）神经网络学习算法作了简单介绍。从SOFM神经网络学习算法的基本思想出发，通过研究SOFM学习算法在设计矢量码书中存在的问题，提出了一种改进算法。最后把这种算法应用在口电话语音压缩编码的参数矢量量化上。计算机仿真结果表明，SOFM神经网络是一种训练语音码书的好工具，改进的SOFM学习算法能够大大减少训练时间，提高整个系统的性能。

Speech Compression Coding Design of IP Phone Based on SOFM Neural Network

The learning algorithm of self-organizing feature map(SOFM) neural network is introduced.According to the basic ideas of the learning algorithm of SOFM neural network,by means of researching the problems existing in SOFM learning algorithm in designing vector codebooks,a modified SOFM learning algorithm is proposed and this learning algorithm is finally used to parameter vector quantitation of speech compression coding of IP phone.The results of computer simulation show that SOFM neural network is a good tool for training speech codebooks,and the modified SOFM learning algorithm can greatly reduce the training time of codebooks and improve the performance of the system.