Link to Colab to view results -> https://colab.research.google.com/drive/17pBWrq-aA8m0Kt9C28dltS9l1sPQeMDd#scrollTo=Q5Kw-HTWq9J5&line=6&uniqifier=1
One can see that in the test dataset which was completely hidden while training that model has learned to remove the noise accurately even for just 3 layers deep dense network, with more layers we can easily attend state of the art result for noise correction.