基于分段的实时声频检索方法  

Real-time frequency retrieval method based on segmentation

作　　者：郑贵滨[1] 韩纪庆[1] 李海峰[1] 郑铁然[1] 

机构地区：[1]哈尔滨工业大学计算机科学与技术学院,哈尔滨150001

出　　处：《声学学报》2006年第2期101-108,共8页Acta Acustica

基　　金：国家自然科学基金(60575030)教育部跨世纪优秀人才培养计划基金哈尔滨市重点科技攻关基金(2005AA1CG036)资助项目

摘　　要：提出了基于分段的实时声频检索方法,并讨论了在实时检索中的控制策略。该方法将检索目标划分为片段序列,并使用检索窗控制参与检索的片段。在多目标检索中,利用声频的类别信息加快检索速度。实验证明检索方法的速度快、可控性好、实时性强,具有良好的缺失鲁棒性(Robustness),查全率和查准率分别达到100％和99．7％;将声频分类可有效提高多目标检索的速度,声频分类方法的平均正确率为95．7％。解决了声频检索中检索反应滞后时间长、检索速度随检索目标长度增加呈线性下降等问题。This paper presents a segmentation-based real-time frequency retrieval method and discusses its control strategy. In the method, the retrieval target is divided into a series of segments and a retrieval window is used to control the search of segments. This method can obtain a very high retrieval speed that is independent on target length and can be controlled by the size of retrieval window. In multi-target retrieval, audio classification is used to reduce the computation of similarity, Experimental results show that recall rate and accuracy rate of retrieval method can achieve 100% and 99.7% respectively and the method can maintain high performance even if large part of the target is absent in the input stream. Classification can effectively improve the speed of search and the average accuracy of classification is 95.7%. Retrieval response can be triggered with little time lag, It is suitable for real-time application to retrieve audio information of any length from unknown input data.

关 键 词：检索方法 实时性 声频 分段 检索速度 控制策略 分类方法 实验证明 滞后时间 目标长度 

分 类 号：TN912.3[电子电信—通信与信息系统]