Detecting Intelligibility by Linear Dimensionality Reduction and Normalized Voice Quality Hierarchical Features

Voice disorders could increase unhealthy social behavior and voice abuse, and dramatically affect the patients’ quality of life. Therefore, automatic intelligibility detection of pathological voices has an important role in the opportune treatment of pathological voices. This paper aims at designing an intelligibility detection system which is characterized by two aspects. First, the system is based on features inspired from voice pathology such as voice quality features, spectral and harmonicity features, and hierarchical features. Second, the intelligibility detection is based on fusion of the individual linear dimensionality reductions such as asymmetric sparse partial least squares (ASPLS) trained by different sets of normalized features. Experimental results show that our method achieves accuracy of 71.88% on the unweighted recall value on the test set, an improvement of 2.98% absolute (4.33% relative) gain over the baseline model accuracy of 68.9%.