Voice Pathology Assessment based on a Dialogue System and Speech Analysis

A system of remotely detecting vocal fold pathologies using telephone quality speech recorded during a telephone dialogue is presented. This study aims at developing a dialogue system using VoiceXML for remote diagnosis of voice pathology. To assess the accuracy of the system, a database of 631 clean speech files of the sustained phonation of the vowel sound /a/ (58 normal subjects, 573 pathologic) from the Disordered Voice Database Model 4337 was transmitted over telephone channels to produce a test corpus. Pitch perturbation features, amplitude perturbation features and a set of measures of the harmonicto-noise ratio are extracted from the clean and transmitted speech files. These feature sets are used to test and train automatic classifiers, employing the method of Linear Discriminant Analysis. Cross-fold validation was employed to measure classifier performances. While a sustained phonation can be classified as normal or pathologic with accuracy greater than 90%, results indicate that a telephone quality speech can be classified as normal or pathologic with an accuracy of 74.15%. Amplitude perturbation features proving most robust in channel transmission. This study highlights the real possibility for remote diagnosis of voice pathology.