Partial discharge (PD) monitoring is one of the diagnostic technique adopted for identifying the variety of defects in transformer insulation. Ultra high frequency (UHF) technique is gaining importance in PD monitoring applications of transformer due to various advantages. Different type of incipient discharges arose from defects in transformer insulation that needs to be identified. In an actual test site there can be noises that can hinder data acquisition and defect identification can become difficult. By using artificially reconstructed signals of known practically occurring defect models, the loss in data can be overcome. In the present study, Deep Convolutional Generative Adversarial Networks (DCGAN) technique is adopted to reconstruct the UHF partial discharge signals with high fidelity. Time-Frequency characteristics of the signals were used to build the DCGAN network and the reconstructed UHF signals are evaluated by studying the frequency characteristics of the generated signal.