This study uses the AutoML tools, machine learning (ML), and deep learning (DL) techniques to classify sounds from images created by Mel spectrograms. Throughout this study, data augmentation techniques were used to strengthen the dataset and thus obtain better results, and then data preprocessing techniques were used to convert the audio files of the GTZAN dataset into spectrograms to extract the most relevant features. The effectiveness of deep neural networks, such as Xception, is highlighted, especially with the application of data augmentation to improve the quality of the model obtained. The results indicate an initial accuracy of 74.84% with the base dataset; however, after the data is augmented, the accuracy significantly increases to 97.47%. Modern methodologies such as convolutional networks and recurrent networks surpass traditional methodologies for music genre classification, underlining the importance of AutoML as a powerful and accessible tool that improves the accuracy of complex classification tasks such as sound classification.