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.
出处
《电子学报》
EI
CAS
CSCD
北大核心
2005年第10期1790-1793,共4页
Acta Electronica Sinica
关键词
连续语音识别
高维空间点覆盖
非特定人语音识别
continuous speech recognition
high-dimensional space covering
speaker independent