The proliferating mortality rate due to Congestive Heart Failure (CHF) has become an area of concern worldwide. The triumph to detect CHF at the earliest possible stage has witnessed various detection methods since its breakthrough. Moreover, Artificial Intelligence has played a pivotal role in the detection process through the boon of technology. Various methods involving statistical, machine, and deep learning have already been discovered, capable of performing accurately. Nevertheless, all the existing methods require the user to perform some manual tasks for processing, or they may not function properly under the deficit of a large dataset. To tide over the problem, we have developed 'HeartGAN .'The 1-Dimensional time series electrocardiogram signal (ECG) has been converted to a 2-Dimensional spectrogram to enhance the model's performance. Those spectrograms are fed into the unsupervised Generative Adversarial Network (GAN), which was self-tailored to augment the dataset up to 30 times with an enormous accuracy of 95.532%. The augmented dataset was then used for autonomous detection of severe CHF with an accuracy of 94.114%. The capability to perform accurately, with minimal data and manual involvement, marks the excellence of our proposed model. Our study can be of immense use to the medical community for severe CHF detection during an emergency. Furthermore, our study can act as a resource to the researchers in the Computational Sciences, Biomedical, and medical arenas, for developing similar technical tools that can aid in saving lives.