Automatic Assessment of Pathological Voice Quality Using Multidimensional Acoustic Analysis Based on the GRBAS Scale

Despite the fact that perceptual evaluation is considered as a gold standard for assessing pathological voice quality, the considerably high inter- and intra-listeners variability associated with different perceptual ratings cannot be ignored. This is probably due to other confounding factors such as listeners’ perceptual bias, listeners’ experience and type of rating scale being used. Automatic objective assessment can serve as a useful tool for diagnosis of pathological voices. Acoustic analysis can be useful in determining severity of dysphonia. The present study aimed to develop a complementary automatic voice assessment system by using multidimensional acoustical measures based on the well-known GRBAS perceptual rating scale. A total of 65 dimensionality measures including traditional acoustic methods, MFCC, Glottal-to-Noise Excitation Methods and nonlinear dynamical analysis were used to compose a matrix of features. To reduce redundancy in features, four different feature extraction techniques were applied. The multiclass classification was carried out by means of RBF kernel-SVM and Extreme Learning Machine. The classification results were moderately correlated with GRBAS ratings of severity, with the best accuracy around 77.55 and 80.58 %, respectively. This suggests that such multidimensional acoustic analysis can be an appropriate assessment tool in determining the presence and severity of voice disorders.