This study introduces a new corpus designed to explore spoken topics in Arabic. Despite the growing interest in building new resources for Arabic, to our knowledge, there is no existing corpus designed specifically for recognizing the topics discussed. We assembled a substantial collection of Arabic podcasts, known as ArPod2.0, covering diverse subjects such as Business, Health, and Technology. This corpus comprises over 30 hours of speech, featuring both standard Arabic and various dialects from regions including Saudi Arabia and Egypt. Careful attention was given to curating this dataset to facilitate computer-based analysis of speech. Our focus was on six main topics, and we employed advanced computer techniques to analyze the data. Our findings highlight the efficacy of one such technique, the Multilayer Perceptron (MLP), which demonstrated exceptional performance in accurately discerning topics, even amidst mixed content. Updated Version