Environmental sound classification is a computational task that belongs to the branch of artificial intelligence called sound recognition. Several techniques and different approaches exist to tackle this task; one that yields excellent results is through the utilization of deep learning techniques, i.e., neural networks. Despite their good results, neural networks in some cases fail to generalize well to new data when the amount of training data is scarce. This can lead to a phenomena called overfitting. A solution to this inconvenience is based on the use of deep generative models to generate synthetic data through the approximation of high-dimensional probability distributions. This allow to generate new samples, similar to the ones used to train the generative model. Generative Adversarial Networks (GANs) are a kind of generative model which trains two neural networks simultaneously in an adversarial way, i.e., pitting one against the other. In this work it is shown the effect of using GANs as data augmentation technique that could be used to improve the performance of different sound classification models.