Abstract:
Automatic speech recognition refers to the process through which speech is converted into text. The best systems for English have achieved a single-digit word error rate (WER) and in some conversational tasks, performance is comparable to human transcribers. Unlike English, speech recognition in Arabic faces many challenges, even with such advanced techniques. Arabic poses a set of unique challenges due to its rich dialectal variety, with modern standard Arabic (MSA) being the only standardized dialect. An objective comparison of the varieties of Arabic dialects could potentially lead to the conclusion that Arabic dialects are historically related, and that they are not mutu ally intelligible languages like English and Dutch. There have been numerous efforts to produce spoken Arabic data set resources. One of the main challenges of processing dialectal speech is to first identify the dialect of the spoken content.
