用于连续语音识别的RBF-Gamma-HMM组合模型

本文提供了一个有特色的、易扩展的多模块ＲＢＦ－Ｇａｍｍａ神经网与ＨＭＭ组合的连续语音识别模型，兼有ＲＢＦ网表达音元空间、Ｇａｍｍａ综合时序相关信息、ＨＭＭ作音元时间集成和扩展等功能，以实现功能互充本模型为基础，将本文提出的各咎改进分类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％。

A RBF-Gamma-HMM Combined Model for Continuous Speech Recognition

A continuous speech recognition model integrated by multi module RBF Gamma neural networks and HMMs is proposed in this paper.In this model,the abilities of RBF Gamma nets that effectively represent the space of speech units and synthesize the temporal correlation information of speech sequence are combined with the abilities of HMMs that integrate and expand the speech units in time domain.They are mutually complementary in function and improve the recognition accuracy obviously.A speaker dependent continuous digits recognition system is realized according to this model and using the learning algorithms proposed in this paper for improving classification performance.The tested digit accuracy is 98.9% and the string accuracy is 94.8%.