基于神经网络的自学习非特定人语音识别研究  

Study of self-learning and speaker-independent speech recognition based on neural networks

作　　者：徐秀平[1] 李柱峰[2] 

机构地区：[1]五邑大学信息学院,广东江门529020 [2]五邑大学物理实验中心,广东江门29020

出　　处：《电声技术》2004年第6期30-32,共3页Audio Engineering

基　　金：江门市科学研究基金资助(20030060);五邑大学科学研究基金资助(20020001).

摘　　要：详细介绍一种基于神经网络的自学习非特定人语音识别方法,首次介绍一种语音识别知识的自动检验方法——LVV法,给出系统原理图和知识库的自动完善原理;介绍一种LEA判别法,实现梯度牛顿有效结合神经网络快速学习方法,并给出了实验结果。The study of self-learning and speaker-independent speech recognition based on neural networks is given. Firstly, the automatic verification method of speech recognition knowledge-LVV is introduced. Then the system principle diagram and automatic adjusting method of knowledge library are presented. The LEA decision method has been given so that grads methods and Newton methods can be effectively combined in neural networks learning. Experimental results are also shown.

关 键 词：神经网络 语音识别 自学习 LVV法 非特定人 

分 类 号：TN912.34[电子电信—通信与信息系统] TP183[电子电信—信息与通信工程]