An Intelligent System for the Diagnosis of Voice Pathology Based on Adversarial Pathological Response (APR) Net Deep Learning Model: An Intelligent System for the Diagnosis of Voice Pathology-Based Deep Learning

1 Introduction

The speech signal is a convolution between the vocal filter and the source signal originates from the lungs. Usually, an indicative breathy phonation in the voice is prominently noticed, which in most cases have been seen with difficulty in swallowing, shortness of breath and mild cough. However, the condition can be reversed after seeking counsel and treatment from expert (Steffen et al., 2011). The pitch of a voice originates due to the periodic opening and closing of the vocal folds. The vocal fold and vocal tract shapes the sound originating from the lungs, which produces the sound that we hear which may be either normal or distorted. It plays a key role in the production of glottal source excitation, and thus any deviation from the normal stature of such signal can indicate the presence of voice pathology in the speaker. Authors of (Gómez-Vilda et al., 2009), (Drugman et al., 2009) have investigated the application of such glottal source excitation signal for detecting pathologies in a voice signal. The mucosal wave spectrum, glottal formant frequency, spectral balance, bandwidth, and average glottal source dynamics, etc. features are popularly used. Advanced deep learning models can be built which can be used with such features to give a very high classification rate. Several studies have shown voice pathology classification using conventional machine learning classifiers but recently, deep learning techniques have evolved.