@RobinHGWells

Used to be the case.

Now it is more of an arms race. Denoising autoencoders can be trained to remove this stuff. And neural networks can be trained on those neural networks to make noise that will fool the autoencoder part, etc... One large GAN.