自然音节状态下声母区别特征检测

从基于言语知觉特性的声母区别特征出发，观察MFCC和美尔能量取值在区别特征有无上的表现差异，修正了声母区别特征的客观参数特性，与修正前的结果相比更合理。定义含有区别特征参数特性的语音帧为有效帧，在自然音节中通过能量确定出声母初判范围位置的基础上，利用该范围内有效帧的含量Para作为检测探针，实现自然音节状态下声母区别特征检测。正检率在80%左右。

Detection of The Mandarin Initials Distinctive Feature in Natural Syllables

According to Mandarin initials distinctive feature systems based on speech perception character-istics,objective signal parameters was corrected which correspond to the distinctive features. Initial dis-tinctive feature was detected tin Natural syllables by parameter Para,which is based on the percentage of effective frames in the initial preliminary position. The effective frame is that speech frame containing the parameter characteristics of a distinctive feature. Initial preliminary position is determined due to energy in Natural syllables. The average detection accuracy of initial distinctive feature was about 80%.