一种新型的动态时间弯曲方法  

A Novel DTW

作　　者：钱小军[1] 周建辉[1] 潘雪增[1] 平玲娣[1] 

机构地区：[1]浙江大学计算机科学和工程学系,杭州310027

出　　处：《计算机科学》2002年第9期69-70,62,共3页Computer Science

摘　　要：1.引语 在语音识别中有一个关键的问题就是语音模式如何比较以确定它们的相似性(或者称为等同性即模式之间的距离),持别在我们应用基于模板的语音识别方法的时候.在基于模板的系统中,如果从频谱对的差别或者说扭曲的角度来看,那么所使用的距离尺度有非常明显的物理意义.A key question in speech recognition is DTW. With the aims of solving the problem of dealing with the similar voices in classical DTW , this paper proposes a new DTW to solve the problem . The new DTW defines the global pattern dissimilarity as the maximum of the distances over all the possible paths. To make the new DTW meaningful, we must impose the constraints on it . On normal circumstances, the constraints include endpoint constraints , monotonicity constraints, local continuity constraints, weighting.At last, this paper points out the significance of the new DTW in the whole process of the speech recognition by showing the diagram of the process.

关 键 词：语音识别 模板识别 动态时间弯曲方法 语言信号处理 语音模式 

分 类 号：TN912.34[电子电信—通信与信息系统]