Development of an Arabic text-to-speech system

Abstract

The field of speech synthesis or Text-To-Speech (TTS) has rapidly expanded during the last few years due to the wide range of applications that require human-machine interaction. Arabic language, being the fourth most spoken language on the globe, has received the attention of the researchers in development of an intelligible and close to natural Text-To-Speech system. Most of the available Arabic TTS Systems have drawbacks such as discontinuity, limited vocabulary and poor naturalness. A rule-based Arabic Text-To-Speech system using hybrid synthesis is presented in this dissertation. Sinusoidal model and concatenation using phonemes as the unit of speech were used to implement the hybrid synthesis system. A set of rules was constructed to achieve letter-to-sound mapping and an exception lexicon to cover the words that does not follow a certain pronunciation rule. Using rule-based synthesis makes the system vocabulary independent. A special phonemes inventory was constructed to be used for the Arabic TTS. Applying the accurate stress pattern for Arabic words was successfully achieved which contributed to the overall quality of the system. The evaluation results shows high level of intelligibility with acceptable naturalness with an overall rate of categorical estimation is 3.458 out of 5.