An Arabic Text To Speech Based on Semi-Syllable Concatenation

Abstract

In the present work, we present our Arabic Semi-Syllable Synthesizer. The work consists of four steps: building a semi-syllable corpus for Arabic Text to Speech, text to phoneme conversion using Arabic Transcription from Orthographic to Phonemes, followed by phoneme to semi-syllables mapping using syllabification rules, and finally speech semi-syllable concatenation. The syllabification step is very important to decide on syllable type and the corresponding stress. The syllabification subsystem was tested on textual data set having more than 1500 words, given by a linguist, especially built for this purpose. The set of words covers all the syllable types in Standard Arabic and all the possible syllable counts in words, ranging from one to seven syllables. The intelligibility, comprehension and naturalness of the synthesized speech were tested using a corpus of 21 Arabic Semantically Unpredictable Sentences and a story of 400 sentences. The results show accurate syllabification and very good quality synthetic speech when compared to other available synthesizers.