面向普通话辅音检测的区别特征参数测量

Distinctive parameter survey of mandarin consonants for speech evaluation

以建立基于特征参数的解析化的普通话辅音发音检测方法为应用目的。根据普通话辅音按发音方式和发音部位分类的特点并结合区别特征理论的二元对立思想,首先提取分析了普通话21个辅音的美尔倒谱系数MFCC和美尔滤波器能量这两类特征,并进一步得到能区别发音方式或发音部位的区别性特征参数k1~k11。在此基础上构建了面向普通话辅音检测的二元分类决策树。与基于HTK的分类结果比较表明：使用基于区别性特征参数的决策树判决的方法对辅音进行分类检测和识别的结果比较稳定,准确率大多在80%以上且有更好的鲁棒性。

This paper is aimed at build a mandarin consonants recognition system based on distinctive parameters.Ac-cording to the classification method of consonants based on articulatory position and manner in combination with bi-nary opposition of distinctive features theory,the two types of features： Mel frequency cepstrum coefficients（MFCC） and Mel-filter energy are analyzed,and then distinctive parameters k1~k11corresponding to articulatory position and manner are further derived.A decision tree based on these parameters is built for classification of consonants.The con-sonants can be recognized by a binary search process from the decision tree.The comparison with HTK results shows that the recognition accuracy of most mandarin consonants based on distinctive parameters is over 80% and has a bet-ter robustness.