基于神经计算机并行运算的连续语音识别  

Research on Continuous Speech Recognition Based on Artificial Neural Computer

作　　者：周旭东[1] 赵鹤鸣[1] 顾晓东[2] 

机构地区：[1]苏州大学通信与电子工程系,苏州215021 [2]北京大学电子学系,北京100871

出　　处：《计算机工程与应用》2002年第16期83-85,共3页Computer Engineering and Applications

摘　　要：人工神经网络具有强大的模式分类能力,除此之外还有一个优点即其大量的并行神经计算能力,使得以低廉的硬件开销来实现高速并行运算成为可能,这在实时模式识别中是非常重要的。文章介绍了以预言神一号小型神经计算机的并行神经运算为核心的连续语音的实时识别的算法,实验表明该算法有很高的实时性和较好的识别率。Artificial Neural Network has strong capability for pattern classification.Besides,another important advantage of ANN is its large-scale parallel neural computing,which makes high-speed low-cost parallel computing possible.This paper presents a speech recognition system which takes advantage of high speed parallel computing ability of Cassan-dra-I neurocomputer.The result shows that this method can realize real time continuous speech recognition with reason-able good correct rate.

关 键 词：神经计算机 并行运算 连续语音识别 人工神经网络 

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]