Detection of Voice Pathologies and Evaluation of Pronunciation Based on Prosodic Features: Case of Arabic Discourse

摘要

Automatic speech processing is a growing area of research, taking advantage of the development and popularity of human-machine interaction applications. Automatic speech recognition is a famous application that allows translating spoken language into text. However, these systems often work in a less efficient way when the produced speech presents degradation. In this article, we address the issue of detecting vocal pathologies contained in Arabic discourse, based on prosodic parameters [12]; we use metrics related to duration describing the rhythm of pronunciation proper to concerned speakers. The obtained results are satisfactory. Indeed, the proposed detector of voice pathologies has attained a performance of 88.63%. Consequently, researchers in similar areas can benefit from our contribution to the development of their systems.