中文摘要：

在分析比较语音共振峰、LPCC、MFCC与Bark谱等语音特征参数的基础上,选择了Bark谱作为普通话水平客观测试中的韵母评测参数,实验结果表明其语音区分能力优于其他特征.根据音征互载的特点,提出了声韵结合的韵母评测建模方法.针对鼻韵母评测精度低的问题,提出了对鼻韵尾段单独建模的二次评价策略.通过对在普通话水平测评现场录制的语音数据进行的评测实验,证明所提出方法能有效提高评测精度,其韵母总体评测精度达到95.80%.

英文摘要：

After analyzing and comparing four kinds of speech features： formant,LPCC,MFCC and Bark spectrum,the last kind is selected as the features of speech in objective evaluation of final,experimental results show that these features are distinct prior in distinguishing the final speech.Depending on the features of Intial-Final Coarticulation,a modeling method of Intial-Final combinations has been proposed.Using the separate models of the nasal segments in the Final,this paper has proposed a secondary evaluation strategy.The evaluation experiments on the speech data recorded in the real Putonghua Proficiency Testing sites show that the proposed methods can efficiently increase the evaluation accuracy.The accuracy of total final objective evaluation is 95.80%.