Today, automatic emotion recognition in speech is one of the most important areas of research in signal processing. Identifying emotional content in Arabic speech is regarded as a very challenging and intricate task due to several obstacles, such as the wide range of cultures and dialects, the influence of cultural factors on emotional expression, and the scarcity of available datasets. This study used a variety of artificial intelligence models, including Xgboost, Adaboost, KNN, DT, and SOM, and a deep-learning model named SERDNN. ANAD was employed as a training dataset, which contains three emotions, "angry", "happy", and "surprised", with 844 features. This study aimed to present a more efficient and accurate technique for recognizing emotions in Arabic speech. Precision, accuracy, recall, and F1-score metrics were utilized to evaluate the effectiveness of the proposed techniques. The results showed that the Xgboost, SOM, and KNN classifiers achieved superior performance in recognizing emotions in Arabic speech. The SERDNN deep learning model outperformed the other techniques, achieving the highest accuracy of 97.40% with a loss rate of 0.1457. Therefore, it can be relied upon and deployed to recognize emotions in Arabic speech.