一种基于高维空间覆盖动态搜索方法的非特定人连续数字语音识别的研究  

Research on Speaker-Independent Continuous Figure Speech-Recognition Based on High-Dimensional Space Covering and Dynamic Scanning

作　　者：王守觉[1] 潘晓霞[2] 徐春燕[2] 陈旭[1] 安冬[1] 曹文明[2] 

机构地区：[1]中国科学院半导体研究所,北京100083 [2]浙江工业大学智能信息系统研究所,浙江杭州310014

出　　处：《电子学报》2005年第10期1790-1793,共4页Acta Electronica Sinica

摘　　要：本文使用高维空间点分布分析原理,在仿生模式识别高维空间点覆盖原理的基础上,提出了一种基于高维空间点覆盖动态搜索理论的非特定人连续数字语音识别的新算法,这种算法可以不经过端点检测和分割,通过对被识别连续数字语音直接进行动态搜索,得到被识别语音到各类高维空间覆盖范围的距离随时间变化曲线,通过距离曲线上的极小值点进行识别.Through analyzing samples distribution in high dimensional space, a novel algorithm for speaker-independent continuous figure speech recognition is presented based on high-dimensional space covering, biomimetic pattern recognition and dynamic scanning in this paper. Without endpoints detection and segmenting the continuous speech is dynamic scanned, then distance curves from the continuous speech to every kind of covering area in high dimensional space are obtained. The continuous speech is recognized by detecting the minimum values in the curves.

关 键 词：连续语音识别 高维空间点覆盖 非特定人语音识别 

分 类 号：TN912.34[电子电信—通信与信息系统]