一种新颖的混合语音检索算法  

Novel mixing speech retrieval algorithm

作　　者：王让定[1] 袁旭海[1] 徐霁[1] 

机构地区：[1]宁波大学纵横智能软件研究所,浙江宁波315211

出　　处：《计算机应用研究》2008年第5期1349-1351,共3页Application Research of Computers

基　　金：国家自然科学基金资助项目(60672070);宁波市科技攻关资助项目(2006B100067)

摘　　要：提出了一种新颖的混合语音检索算法。利用ICA的盲分离特性,将包含两个人声音的混合语音分离成只含一个人的单一语音,以分离出来的单一语音为新样本建立新的语音库,并与基于混合语音的语音库建立链接关系。查询时,利用DTW技术进行匹配检索,实现了基于混合语音的检索功能。实验表明该方法能够在混合语音中查询到用户所要求的记录。A novel mixing speech retrieval algorithm was presented. The characteristics of BBS （ blind source separation） of ICA （independent component analysis） was used. The mixing speech records which contained a couple of different person＇ s voice were separated into single speech records which contained only one person＇ s voice. New speech database based on single speech records was established. The linkage between new speech database and original speech database which contained mixinged speech records was established. In query phrase, DTW （dynamic time warping） algorithm was used for retrieval matching and the function of retrieval based on mixing speech was achieved. Experimental result shows that algorithm of this paper can find demand records in mixing speech records.

关 键 词：混合语音 语音检索 独立分量分析 MEL频率倒谱系数 动态时间归整 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]