提取耳蜗熵的语音可懂度评价算法  

CSE-based prediction of speech intelligibility

作　　者：付建梅[1] 张雪英[1] 曹棣[2] 马建芬[2] 

机构地区：[1]太原理工大学信息工程学院,太原030024 [2]太原理工大学计算机科学与技术学院,太原030024

出　　处：《中国科技论文》2013年第10期1017-1020,共4页China Sciencepaper

基　　金：高等学校博士学科点专项科研基金资助项目(20111402110013)

摘　　要：为了提高主观和客观听力条件下语音可懂度的相关度,提出用耳蜗熵值来衡量语音的可懂度。利用提取的耳蜗熵值分段后的语音作为归一化协方差(normalized covariance measure,NCM)的语音可懂度算法的输入,分析高熵值语音段和低熵值语音段主客观相关系数的高低。结果表明,提取的高耳蜗熵值语音段比低熵值语音段对语音可懂度算法与主观评价的相关性高,高熵值语音段携带有更多的语音信息。To improve the correlation of speech intelligibility between the subjective and objective audiometry conditions, the ap- proach for measuring speech intelligibility with cochlea-scaled entropy is proposed. The extracted cochlea entropy segmented with the speech-intelligibility algorithms of normalized covariance measure （NCM） works as the input, the subjective and objective cor- relation coefficients of high and low entropy speech segments are analyzed. The results indicate that the correlation in the high en- tropy segments is higher than that of the low entropy segments, and high-entropy speech segments carry more speech information.

关 键 词：语音增强 语音可懂度 耳蜗熵值 归一化协方差 

分 类 号：TN912.3[电子电信—通信与信息系统]