Just when I thought I had come across a great idea, Wikipedia told me that I'm not the first to think about this: the green one) could have actually happened in the lapse of time we don't have info about. The recovered signal using the stated method would be the blue one, while in fact any signal (e.g. The signal could be anything but a peak: it has time to do lots of things until the clipping stops.

On the other hand, if the signal is heavily clipped, the chance of guessing it right is unarguably lower. Hopefully, the result will be better than the clipped version. Of course, you have a probability of guessing how the original signal actually was, so you will probably add some distortion. An statistical approach would probably work, but it is not a trivial task. This kind of method could detect the clipped intervals and ''soften'' them, based in the previous behaviour of the signal.

Clipping is present only in the greatest peaks, and therefore appears for a short time lapse. Let me explain: think about a softly clipped signal. I think it is possible, but depends on how the signal is clipped.

How to fix clipping in audacity