基于LPCC和MFCC参数的病理嗓音识别研究  

Identification of Pathological Voices based on LPCC and MFCC

作　　者：莫丽花[1] 周孝进[1] 张晓俊[1] 陶智[1] 赵鹤鸣[2] 顾济华[1] 

机构地区：[1]苏州大学物理科学与技术学院,江苏苏州215006 [2]苏州大学电子信息学院,江苏苏州215006

出　　处：《通信技术》2012年第1期87-89,共3页Communications Technology

摘　　要：自动检测正常嗓音和病理嗓音的关键是选出有效的特征参数,并对其进行优化得到简单易实现的参数。同时选择合适的识别模型对正常嗓音和病理嗓音进行识别以得到最好的识别率。为了能实时、便利地检测正常嗓音和病理嗓音,这里提出了线性预测倒谱系数(LPCC)和MEL频率倒谱系数(MFCC)声学特征参数,采用动态时间规整(DTW)算法进行识别,实验结果表明该模型的识别率可达到90%以上,且MFCC方法优于LPCC。The choice of effective feature parameters is the key in the automatic detection of normal and pathological voices, then these parameters are optimized to obtain easy and achievable parameters. Meanwhile, the selection of an appropriate model could achieve the best recognition rate. In order to detect normal and pathological voices instantly and conveniently, linear prediction cepstral coefficients （ LPCC ） and Mel cepstral coefficients （ MFCC ） are proposed to identify them respectively by using dynamic time warping （ DTW ） model. The experimental results show that the correct recognition rate could reach 90%, and MFCC method is better than LPCC.

关 键 词：病理嗓音识别 线性预测倒谱系数 MEL频率倒谱系数 动态时间规整 

分 类 号：TN912.34[电子电信—通信与信息系统]