语音倾向性分析中的特征抽取研究  

Research of feature extraction in voice orientation analysis

作　　者：魏平杰 樊兴华[1] 

机构地区：[1]重庆邮电大学中文信息处理研究所计算智能重庆市重点实验室,重庆400065

出　　处：《计算机应用研究》2014年第12期3580-3583,共4页Application Research of Computers

摘　　要：提出将声学特征与语义特征相结合来判断语音倾向性的方法。首先从语音中分别提取语义特征及声学特征,然后将语义特征与声学特征进行组合,最后用基于SVM的两步分类方法进行训练和识别。分析和比较了常用的特征及组合,发现将语义特征与声学特征结合起来后效果明显,比单独使用语义特征最高能提高3%,比单独用声学特征的识别率最高能提高14%。This paper presented a new method of combining of semantic features with acoustic features to determine speech orientation. Firstly,it extracted the semantic features and acoustic features from the speech. Secondly,it combined these two types of features. Finally,it used the SVM method which based on two-steps classification to train and identify. This paper analyzed and compared the commonly used features and compositions,found that semantic features combined with the acoustic features,the recognition rate is highestly improved by 3% and 14% respectively which compared with the method of using semantic features and voice acoustic features.

关 键 词：倾向性分析 语音倾向性 声学特征 语义特征 

分 类 号：TP391.42[自动化与计算机技术—计算机应用技术]