摘要
提出了一种基于分类回归树(Classification And Regression Tree,CART)的汉语韵律短语识别方法。该方法从语音流中提取与韵律短语边界有关的声学特征,从文本中提取短语边界的语言学特征,并将两类特征有机结合构成CART特征集,建立CART决策模型。开放测试结果显示,利用该CART模型在词边界中识别韵律短语边界,其识别准确率平均可达95.91%。
This paper presents a CART-based method for identifying the Chinese prosodic phrase.Firstly,it obtains acoustic characteristics which have relation to the boundary of prosodic phrase from speech,and it gain linguistic characteristics of prosodic phrase boundary from text.Secondly,it combines these characteristics effectively to construct characteristic muster,and then use it to build CART model.The results of opening test show that identifying the boundary of Chinese prosodic phrase using this CART model,its precision can reach 95.91% averagely.
出处
《计算机工程与应用》
CSCD
北大核心
2008年第6期169-171,共3页
Computer Engineering and Applications
基金
国家自然科学基金(the National Natural Science Foundation of Chinaunder Grant No.60572159,No.60573184)。
