Given the immense importance of emotions in human relationships, many researchers have been interested in setting up systems for emotion recognition through speech. These systems provide applications that have a positive impacts on human life. Emotion recognition systems are based on computer science techniques, but fundamental studies on emotions are done by psychologists, who have shown that emotions differ from one society to another. Taking into account these variations, and to enable the construction of systems that can properly recognize emotion in Moore language, we have built the first speech emotional dataset in this language namely Moore Natural Emotions Speech Dataset (MNESD). This is a fundamental step in the implementation of these recognition systems. The corpus contains 6017 audio files on natural emotions collected from the public radio of Burkina Faso. It covers 7 emotions (joy, satisfaction, neutral, anger, contempt, disappointment and sadness). Through this study, we aim to encourage studies on speech emotions recognition in African societies by making available a speech corpus in Moore. We carried out the first experimentations with MNESD combining Mel Frequency Cepstral Coefficients (MFCC) and Fundamental Frequency (F0) coefficients. The best accuracy obtained is 90% using LSTM classifier. The dataset is available on Kaggle, and the link is provided at the conclusion of this paper.