摘要
语音关键词识别是近年来颇受重视的一个研究领域,文章基于特征空间轨迹的时间规整化原理,提出了一种高性能的关键词检测法,并探讨了轨迹等分长度对该算法检测性能的影响。实验结果表明,基于特征空间轨迹匹配方式的关键词检测法的检测性能接近于人工检测,具有一定的实用性。
Keyword spotting(KWS )has been an active research issue in recent years.This paper proposes a new,high-performanced method of keyword spotting based on theory of speech feature space trace time normalization,and tries to find out what effect the trace fragment length has to the performance of this method.Experiments show that this method has achieved performance close to that of manual spotting,having some practicability.
出处
《计算机工程与应用》
CSCD
北大核心
2003年第36期83-86,共4页
Computer Engineering and Applications
基金
上海交通大学985工程计划资助
