摘要
该文提出一种普通话单字发音准确度的计算机辅助评价方法。以RBF网络为评价系统,选用归一化BARK谱为特征,并且用标准发音聚类构成的模板特征作为其它发音的参照。最后将样本特征与模板特征的差值平方矢量作为RBF网络的输入,进行网络的训练和评价。实验表明,该方法取得了较好的效果。
This paper studies briefly a new method of computer aided assessment of mandarin pronouncing accuracy based on RBF Neural Network,It uses normalized BARK as features,select correct pronunciation as reference model,and finally uses the vector of square of difference between the sample feature and the model feature as the input of RBF network.The experimental results showed good performance.
出处
《计算机工程与应用》
CSCD
北大核心
2005年第26期72-74,共3页
Computer Engineering and Applications
