Sentiment analysis is used in several fields, such as teleconsultations in the medical field, or emergency calls to the police or firefighters. Voice analysis is used to know the emotional state of the person on the device. In our study, we propose an automated system to classify Moroccan dialect speakers’ emotion using Deep Learning (DL) and voice-extracted features such as Mel-Frequency Cepstral Coefficients (MFCC). Moroccan dialect also known by the name ‘Darija’, is the dialect spoken by most citizens of Morocco. The Darija is very different from Arabic official dialect that we have in newspapers and TV. While previous research has predominantly focused on textual sentiment analysis using Natural Language Processing (NLP) techniques, our work introduces an approach by incorporating voice sentiment multiclassification for the Moroccan dialect. In our paper, we propose a voice sentiment analysis classification on a database that we created and that contains 2000audios recorded with Moroccan dialect speakers. The proposed automated system classifies 5 different types of sentiment happy, neutral, sad, angry or fearful.