Automatic speech recognition for dysphonic patients by using oriented local features

The number of patients with voice pathology has increased significantly in recent years. The disability or illness of a person should not deprive him from taking benefits of the technology advances that is changing the daily life. For example, modern day speech recognition technology should be capable to recognize a speech from a normal person as well as a person having dysphonic. In this paper, we propose a new speech feature to use in automatic speech recognition system of disordered speech. We compare the performance of this feature with the most widely used speech feature in speech recognition. The comparison is done using spoken words uttered by both normal and dysphonic patients. The obtained results with the proposed technique are good and comparable to the existing method. Copyright ISCA, CAINE 2014.