【作者】 张全； 张倪； 韦向峰；

【Author】 ZHANG Quan ZHANG Ni WEI XiangfengInstitute of Acoustics, CAS, Beijing 100080 Center of Advance Software Engineering (CASE), CAS, Beijing 100080

【机构】 中国科学院声学研究所； 中国科学院北京软件工程研制中心；

【摘要】 n-gram作为语言处理模型已经广泛应用于语音识别系统中,但处理结果中经常出现错误.本文指明影响语音识别错误产生的主要因素,制定了基于概念层次网络(简称HNC)句类分析技术的纠错处理策略.同时构筑了纠错处理的实验系统.测试表明,该技术能够突破现有统计模型的局限,取得理想的结果,有望成为提高语音识别性能的、新的语言处理模型.更多还原

【Abstract】 The N-Gram is used widely in Speech Recognition Systems, however, the errors are often occurred in the output. In this paper, the main issues which cause the errors are discussed., and the correcting tactics based on the Hierarchical Networks of Concepts (HNC) Analysis of Sentence Category (SCA) technology are described. Meanwhile an experiment system based on the tactics have been built. The result of the system indicated that, the SCA is able to break through the limits of the statistical models, to get a better result. It is hopeful that the SCA will be used as a new linguistic processing model which is able to promote the performance of speech recognition system.更多还原