Identification of Pathological Speech Disorder Using Machine Learning

Abstract

Speech is the method of generating specific sounds that convey meaning to the listener. This ability of a person could be hampered due to several reasons leading to speech disorders. A speech disorder refers to any circumstance that affects a person’s capability to create sounds that produce words. According to a survey conducted by All India Institute of Speech and Hearing, Manasgangotri, Mysore, speech impairment is listed as the fifth-highest occurring disability in India. There are different types of speech disorders such as Stuttering, Apraxia, Dysarthria, Dysphonia, that affect a person’s capability to form sounds that permit him/her to converse with other people. Recently computer-based detection of speech disorders has got attention from researchers and doctors as it can be used to detect speech disorders effectively. A similar approach has been proposed in this paper to identify pathological disorders using Support Vector Machines and Gaussian Mixture Model. It can be beneficial to people in rural areas as there is a scarcity of well-trained speech pathologists. Also, equipments needed to identify speech disorders are expensive. Hence there is a need for an automated system to assist a pathologist in detecting speech disorders. In this work, speech dataset from Saarbruecken, related to the above-mentioned diseases was collected and were classified as healthy and unhealthy. An accuracy of 96% was attained using Gaussian Mixture Model.