In recent years, neural networks are utilized to search for radio technosignatures that are narrow-band, drifting signals in the dynamic spectra (or “waterfall plots''). Due to the rare occurrence of known extraterrestrial artificial signals, one will likely not have sufficient actual data to train the neural network, and synthetic data will therefore be needed. A new way of generating synthetic data is to employ the Generative Adversarial Network (GAN) method, which consists of two networks: the generator network will train to generate new examples, and the discriminator network will try to classify examples as either real or fake. These two models are trained together, competing until the generator network can generate plausible examples. We present a GAN model that generates synthetic waterfall plots that can be used in the training of classification networks. Our approach is developed and tested on clean/noisy spectrogram data that contains constant signal.