语音质量客观评价方法研究及实现  

Research and implementation on objective speech quality evaluation

作　　者：方凡泉[1] 李心广[1] 王桂珍[1] 林江豪[1] 

机构地区：[1]广东外语外贸大学,广东广州510006

出　　处：《广州大学学报（自然科学版）》2011年第1期65-69,共5页Journal of Guangzhou University:Natural Science Edition

基　　金：广东省自然科学基金项目(9151042001000017);广东省科技计划项目(2008B080701007)资助

摘　　要：语音质量的客观评价在语言自主学习中具有重大的意义.文章首先介绍了语音质量客观评价过程中语音信号的预处理单元;其次,介绍了不同的语音特征提取算法,比较选择了更符合人耳听觉模型的MFCC特征,并给出特征提取过程及结果;最后,比较当前评价模型(DTW和HMM)的优缺点,并提出了采用HMM模型进行评价的方法,设计系统验证了该方法下评价的客观性.Objective speech quality evaluation plays a pivotal role in SALL （Self-Access Language Learning）. This thesis firstly introduces the pre-treatment unit of voice signals in the process of objective speech quality evaluation. Then, it also presents diverse voice features extraction algorithms and makes comparisons between them. After that, the MFCC feature, which is closer to human auditory model,is chosen in the extraction process. Meanwhile, the whole process and its results are provided. Finally, we compare the advantages and disadvantages of current evaluation models （DTW and HMM）, and propose an evaluation method which adopts the HMM model, and thus design a system to test and verify the objectivity of evaluation using this method.

关 键 词：语音质量 客观评价 HMM模型 语音特征 

分 类 号：H01[语言文字—语言学]