本文提供了一个有特色的、易扩展的多模块 Ｒ Ｂ Ｆ Ｇａｍ ｍａ 神经网与 Ｈ Ｍ Ｍ 组合的连续语音识别模型，兼有 Ｒ Ｂ Ｆ 网表达音元空间、 Ｇａｍ ｍａ 网综合时序相关信息、 Ｈ Ｍ Ｍ 作音元时间集成和扩展等功能，以实现功能互补．以本模型为基础，将本文提出的各种改进分类性能的学习算法用于特定人连续数字语音识别，其字正识率达到９８９ ％ ，串正识率达到９４８ ％ ． This article provides a unique and extensible multi-module R B F Gam ma neural network and H M M combination of continuous speech recognition model, both R B F net expression of the acoustic space, Gam ma network comprehensive timing information , H M M for time integration and expansion of audio and other functions, in order to achieve complementary functions. Based on this model, the learning algorithm proposed in this paper, which improves classification performance, is applied to continuous digital speech recognition of specific people. The word recognition rate reaches 98.9% and the string positive recognition rate reaches 94.8%.