Automatic diagnosis of pathological voices

Spasmodic dysphonia (SD) is a voice disorder characterized by voice breaks. Muscle tension dysphonia (MTD) is a form of voice misuse characterized by excessive muscular effort. While the first pathology is not a psychological condition and has a neurological origin, the last one does not include a neurological disorder and is correctable with voice therapy. Patients with SD are often not identified for treatment. These two pathologies are only correctly differentiated by experts. The importance of a correct diagnosis is directly related with the application of the suitable treatment. Our goal is to provide voice pathologists with a new tool to confirm their diagnosis. In the present work, we present a preliminary approach to this problem, building an automatic classifier using acoustical measurements on registered sustained vowels / a / and pattern recognition tools based on neural networks. As long as we know, there are not previous published works in automatic classification of these two pathologies. However, there are works on automatic classification between normal and pathological voices. Our results overcome the best reported classification between pathological and normal voices, and have a good discrimination between SD and MTD.