A phonetic transcription system of Arabic text

Within the framework of developing an unlimited vocabulary speech synthesizer a rule based transcription system has been developed. It requires as an input a partly diacritized Arabic text. The inventory of the output sounds consists of 24 consonants and 16 vocalic allophones. Three kinds of word stress are defined; main stress for the lexeme, secondary for an eventually unstressed long vowel, and tertiary for the suffix cluster. The locations of the first and third stress types are determined, after word syllabification and a tree search of suffix(es), applying the well established stress rules in Arabic. A relatively small dictionary is used for words with exceptional pronounciation. The transcription system is coded in Pascal language consisting of some 950 instructions. On the avarege a processing speed of 65 word/sec is achieved on an IBM-PC-AT.