Analysis, overview and creation of an Arabic LVCSR

Abstract

As the standardized version of the Arabic Language, Modern Standard Arabic (MSA) is the most prevalent form of this language. MSA is also the third most spoken language in the world with over 300 million speakers. Moreover, its history dates back to the eighth century B.C., resulting in a striking rich linguistic structure. This linguistic structure brings along a broad range of challenges in terms of Large Vocabulary Continous Speech Recognition (LVCSR) execution. In this dissertation, we present an analysis on the Modern Standard Arabic language from a linguistic perspective together with the state of the art of the current Arabic LVCSR from the technical perspective by reproducing and evaluating its state of the art.