利用无监督自适应的兴奋解说检测和体育比赛精彩片断提取  

EXCITED COMMENTATOR SPEECH DETECTION WITH UNSUPERVISED ADAPTATION FOR SPORTS HIGHLIGHT EXTRACTION

作　　者：孙怿[1] 欧智坚[1] 胡炜 

机构地区：[1]清华大学电子工程系,北京100084 [2]英特尔中国研究中心,北京100080

出　　处：《计算机应用与软件》2008年第11期20-22,共3页Computer Applications and Software

基　　金：国家863项目资助(2006AA01Z149)

摘　　要：提出一种通过兴奋解说检测进行体育比赛精彩片断提取的方法。该方法包括训练和检测两个阶段:在训练中,基于训练数据对兴奋语音和普通语音分别建立高斯混合模型GMM(Gaussian Mixture Model),构成初始的分类器;在集外检测中,首先使用最大后验方法MAP(Maximum A Posteriori),基于测试数据对初始模型进行无监督自适应,进而利用更新后模型构成分类器识别体育解说的兴奋部分,经进一步处理得到精彩片断。将该方法用于足球比赛视频,实验表明,该方法能够召回87%的进球。引入无监督自适应有效地减少了由干训练数据与测试数据失配造成的性能下降,提高了兴奋解说检测和精彩片段提取的性能。An excited commentator speech detection method with unsupervised adaptation is applied to sports highlight extraction. The excited speech and normal speech are modeled as two Gaussian mixture models （GMMs） ,and they are updated to compensate for the acoustic mismatch between training and test data via Maximum a posteriori（MAP） adaptation. The adaptation is operated in an unsupervised mode,and a first pass of detection using old GMMs is performed to produce hypothesized classification results. Experimental results demonstrate the effectiveness of the proposed approach. Based on the excited speech detection alone,87% of the goal events can be recalled.

关 键 词：体育比赛精彩片断提取 无监督自适应 高斯混合模型 

分 类 号：TP273.2[自动化与计算机技术—检测技术与自动化装置] G80-05[自动化与计算机技术—控制科学与工程]