BAC TTS Corpus: Rich Arabic Database for Speech Synthesis

Recorded Speech Corpus (RSC) is one of the key elements of corpus-based synthesis. Sentence length and corpus size are additional factors that affect the Speech Synthesis (SS) sound quality, along with the quality and coverage of the Speech Corpus (SC) in each unit and its representations. This study suggests a method for creating SS systems based on Arabic corpora. This fact covers the investigation of the standards used to create a corpus-based SS. More than 5 million words have been gathered in an initial corpus in order to accomplish our goal. To this end, a variety of sources are gathered, including newspapers and the Shamela Arab Library. To design the sentences for the SC, the initial corpus was analyzed using phonemes and the occurrence of words to identify all high frequency phonemes and words. A 202-sentence supervised phonetically balanced SC with 6174 phonemes has been created. The findings demonstrate that the unit coverage in the corpus has an actual influence on the perceived SS quality.