基于MATLAB的MFCC在语音相似度评价中的应用  

Application of MFCC Based on MATLAB in Evaluation of Speech Similarity

作　　者：赵桂兰[1] 孙东[2] ZHAO Gui-lan SUN Dong(Department of Foreign Language, Lanzhou Petrochemical College of Vocational Technology, Lanzhou 730060, China Educational Administration Supervision Jiuquan Vocational and Technical College Office, Jiuquan Gansu 735000, China)

机构地区：[1]兰州石化职业技术学院应用外语系,甘肃兰州730060 [2]酒泉职业技术学院教务督导室,甘肃酒泉735000

出　　处：《兰州工业学院学报》2016年第5期51-54,共4页Journal of Lanzhou Institute of Technology

基　　金：甘肃省科技厅项目(1204GKCA004);甘肃省财政厅专项资金立项资助(甘财教[2013]116号);兰州石化职业技术学院规划课题(JY2015-05)

摘　　要：将梅尔(Mel)倒谱系数MFCC作为语音特征相似度的评价参数,利用帧间平均欧氏距离来衡量语音的相似度,结合相似匹配图,定量及定性的给出测试者语音质量优劣的评价结果,评价过程可由MATLAB方便地完成.该评价方法有助于语言学者或模仿者提高发音质量和朗读能力,也有助语音模式识别.实例分析表明该评价方法的评价结果与专家评委的评判结果一致.By using inter frame average Euclidean distance to measure phonological similarity,we combined similar graph matching to test the voice quality evaluation results,thus the evaluation process can be made by MATLAB conveniently. This method is helpful to improve the quality of language pronunciation for scholars or imitators to read as well as speech pattern recognition. The example analysis shows that the results of evaluation is consistent with the experts＇ judgments.

关 键 词：语音相似度 MFCC参数 相似匹配图 评价 

分 类 号：TN912.34[电子电信—通信与信息系统]