Pathological Assessment of Vocal Fold Nodules and Polyp via Fractal Dimension of Patients' Voices

In this paper, we are going to evaluate the role of nonlinear feature, fractal dimension, in discriminating patients with speech disorders from normal subjects. Laryngeal pathologies usually cause an asymmetry in oscillation of vocal folds. This leads to sub-harmonics and chaos in generated voices. In such condition, using nonlinear dynamic features, like fractal dimension, seems to be efficient approach to analyze understudied voice. To calculate fractal dimension of voice sample, we exploit two methods of "Petrosian" and "Katz" and compare them with each other. Moreover, in order to evaluate role of frequency sub-bands in diagnosis process, voice signals are decomposed into two sub- bands by wavelet filter bank and fractal features extracted distinctively for each band. To simplify current work, we only conduct our experiments on discriminating nodule and polyp disorders from normal subjects. The best classification result is obtained 88.9% using combination of fractal dimensions of voices and their sub-bands.