KEDAS: A validated Arabic Speech Emotion Dataset

This paper aims to construct and validate Arabic emotional speech data for speech emotion recognition systems. This work took place in two main phases: collecting, evaluating, and validating audio data according to human perception. Then the data is processed in terms of framing, normalisation, and background denoising to make the data accessible to models of emotion recognition systems. The result of this work is a KasdiI-Merbah Ouargla University Dataset(Kedas Dataset) for emotional speech in standard Arabic language, which includes 5000 audio recording files distributed to emotions of anger, happiness, sadness, fear and neutrality. A total of 10 of the most used phrases in communication are based on literary and scientific studies. Five hundred actors from the university community recorded the data (students, professors and staff), and fifty-five others evaluated the validity of the identification and the reliability, including Arabic linguists, literary researchers, and specialists in clinical psychology.