摘要
本文提出了有限状态径向基函数(FSRBF)网络结构。它可用K-均值聚类算法和最小二乘算法分层独立训练,训练速度快。通过汉语语音识别实验,研究了FSRBF网及子网的特性。结果表明,FSRBF结构很适于处理时序信息,易于推广到其他识别单元的系统中。
In this paper,Finite-State Radial Basis Function (FSRBF) network structure is proposed,which can be trained layer-by-layer by means of k-means clustering and LMS algorithm.The properties of FSRBF net and subnet are studied by Chinese digit recognition experiments. The results suggest that FSRBF is suitable for time series signal processing and can be generalized easily.
出处
《电子学报》
EI
CAS
CSCD
北大核心
1996年第1期101-104,共4页
Acta Electronica Sinica
关键词
径向基函数
语音识别
神经网络
有限状态
Radial basis function,Speech recognition,Neural networks,Finite state