LAUSR

The main obstacle in the development of a robust Arabic text-to-speech system is the ability to place diacritics, that is, small marks above and below the raw text that indicate the correct pronunciation. In this paper, we propose a system that can retrieve the diacritics that match a given Arabic text. First, the system injects the raw text into an engine that is based on a single hidden Markov model. The engine then generates an optimal path through its states. The system finally matches the state sequence with its equivalent diacritics and sets them in place within the text. Experiential results on diverse data sets demonstrate the robustness of the proposed system, even with samples that are novel to the system.