Objective assessment of pathological voice quality

This paper describes a new technique for objective analysis of voice quality in male volunteers (normal) and male patients (abnormal) recovering from cancer of the larynx. The fundamental-harmonic normalised (FHN) spectra is derived from the power spectrum of normal and abnormal phonating the vowel /i/. A mixture of Gaussian are fitted to the FHN spectra and the parameters obtained are the means, standard deviation, and mass of the density functions. A dataset of 116 normal and abnormal is used to train and test a multilayer perceptron (MLP) neural network. The ability of the trained MLP to classify impedance signals as normal or abnormal is presented.