Restoring audio after compression using neural networks looks neat, though I wonder whether it’s anything like the compression used on some photocopiers that substitutes the wrong data and thus changes meaning: auphonic.com/blog/2018…