Abstract
In this paper we describe a new emotional speech corpus recorded for Modern Standard Arabic (MSA). The newly designed corpus contains five target emotions, namely neutral, sadness, happy, surprised, and questioning, and it consists of 16 selected sentences that were read by 20 male and female native Arabic speakers. A human perceptual test was then applied to the recorded corpus. The test was performed by nine additional individuals. The results of the perceptual test verified the correctness of the intended emotions at a rate of 82.13%. Specifically, the most accurately identified emotion was "questioning," while the least identified emotion was "happy." Subsequent analyses of the results revealed that content sentences play an important role in influencing speakers with respect to controlling the intended emotion. This corpus is the first MSA corpus to be built using multiple variables, including gender, language content, and other speaker demographic data.