The electrocardiogram (ECG) is one of the most commonly used tools for assessing various cardiac conditions. This is achieved by analyzing the signal of the electrical activity of the heart, however, these signals are often corrupted by various types of noise, complicating the process of effectively diagnosing patients. This thesis addresses the challenge of denoising ECG signals from common noises like baseline wander (BW), muscular artifacts (MA) and powerline interference (PLI) by introducing a novel deep learning methodology that is based on two key elements: noise estimation and multitask learning (MTL). With noise estimation, the conventional approach of estimating the clean ECG signal from the noisy one is inverted, instead, the model focuses on directly estimating the noises present in the signal.