Try this with a deep convolutional neural network to see if it's possible to at least reduce the sound by a certain percentage.