半导体神经计算机的硬件实现及其在连续语音识别中的应用  

Hardware Realization of Semiconductor Neurocomputer and Its Application to Continuous Speech Recognition

作　　者：王守觉[1] 曹文明[1] 

机构地区：[1]中国科学院半导体研究所神经网络研究室

出　　处：《电子学报》2006年第2期267-271,共5页Acta Electronica Sinica

基　　金：国家自然科学基金(No.60576055)

摘　　要：本文首先分析了以PC机作为宿主机的半导体神经网络处理机CASSANDRA-I,进一步介绍了新的半导体神经计算机CASSANDRA-II的系统实现和功能特性,并将其应用到问候语语音识别中,实验结果表明CASSAN-DRA-II神经计算机识别结果优于HMM模型的识别结果.The first semiconductor nerocomputer CASSANDRA-Ⅰ based on host-PC is analyzed, and the realization and characteristics of the novel semiconductor neurocomputer CASSANDRA-Ⅱ are described. Finally, the semiconductor neurocomputer CASSANDRA-Ⅱ is used to greeting speech recognition, the experiment results show that it provides higher discriminability for keyword candidates than HMM model.

关 键 词：半导体 神经网络 语音识别 

分 类 号：O235[理学—运筹学与控制论]