针对现有汉语重音检测方法正确率较低的问题,利用声学、词典和语法相关特征的不同分类器组合,基于Boosting分类回归树+条件随机场的互补模型,提出一种改进的汉语重音检测方法。在ASCCD语料库上的实验结果表明,该方法能获得84.9%的重音检测正确率,相比基于神经网络+决策树的基线系统提高2.7%。
Aiming at the problem that existing mandarin stress detection method has low accuracy rate,this paper proposes the complementary model method based on Boosting Classification and Regression Tree(CART) and Conditional Random Fields(CRFs).It is the combination of different classifiers to detect Mandarin character stress by using the acoustic,lexical and syntactic related features.In the ASCCD corpus,the complementary model achieves 84.9% stress detection correct rate,and there is 2.7% improvement when compared with the baseline system based on Neural Network(NN) and decision tree.