基于音素的非特定人英语命令词识别算法研究  

Research on an Algorithm of Phoneme based Speaker-independent English Commands Recognition

作　　者：贲俊[1,2] 余小清[1,2] 万旺根[1,2] 

机构地区：[1]上海大学通信与信息工程学院,上海200072 [2]香港理工大学多媒体创新中心,中国香港九龙

出　　处：《信号处理》2002年第6期535-538,共4页Journal of Signal Processing

基　　金：本文得到国家自然科学基金项目(No.60072031);香港理工大学研究基金项目(No.G-T499)的资助

摘　　要：本文提出了一种新的基于音素的非特定人英语命令词识别算法,并在此算法基础上构建了一个非特定人英语命令词识别系统。结合非特定人语音识别系统的特点,系统的实现采用了HTK与VisualC++两种工具混和使用,提高了整个系统的开发效率。在识别阶段将置信度评估和不完全匹配的方法结合在一起,在一定程度上提高了识别的质量,在词汇量大于10的情况下取得了87.8％的识别率。In this paper we propose a new algorithm of phoneme based speaker-independent English commands recognition and develop a speaker-independent English commands recognition system based on it. It can accelerate the whole system development by mixed using HTK and Visual C++ based on character of speaker-independent speech recognition. In recognition phase we combine confidence measures with incomplete matching, which considerably improve the quality of recognition. RA (Recognition accuracy) is 87.8% when the vocabulary is more than 10.

关 键 词：音素 非特定人英语命令词识别算法 语音识别 置信度 

分 类 号：TN912.34[电子电信—通信与信息系统]