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.
出处
《计算机应用研究》
CSCD
北大核心
2014年第12期3580-3583,共4页
Application Research of Computers
关键词
倾向性分析
语音倾向性
声学特征
语义特征
orientation analysis
speech orientation
acoustic features
semantic features
