作者: Fay. Ykhlef , W. Benzaba , R. Boutaleb , Jessus B. Alonso , Far. Ykhlef
DOI: 10.1109/ISM.2015.23
关键词: Normality 、 Computer science 、 Statistical model 、 Pattern recognition 、 Gaussian 、 Linear discriminant analysis 、 Set (abstract data type) 、 Artificial intelligence 、 Mel-frequency cepstrum 、 Feature (machine learning) 、 Reduction (complexity)
摘要: In this paper, we propose another approach for the measurement of degree voice normality based on statistical modeling. The basic methodology behind proposed is "Pathological Likelihood Index" reported by Godino-Llorente JI. et al. [1]. major innovations are: exploring a reduced set Mel frequency cepstral coefficients (MFCC) and ignoring their derivatives, linear projection MFCCs into one dimensional space using Fisher's discriminant, and, modeling build around single Gaussian distributions instead mixtures distributions. We have evaluated Massachusetts Eye Ear Infirmary database (MEEI). obtained results are better than in