In recent years, researchers have been investing intensively in the field of AI and robotics so that these robots can communicate as normally as possible with humans. These robots need to recognize human emotions to understand them and respond to their desires. We can recognize the emotion of the person either by his facial expressions, his voice, the gestures of his body or by his words. In this work we just focused on the speech, we worked with the RAVDESS corpus, which contains 8 emotions (neutral, happy, sad, fear, angry, surprise, calm and disgust). For emotion recognition, we have used deep learning which we tested two different architectures based on CNN and compared their accuracies.