I know all that, and of course the programs I'm talking about have no way of restoring exactly what was there before the clipping. But from what I understand, they guess at it by analyzing what the wave looks like coming into and going out of the clipped sections; there may be some other fancy mathematical transforms going on too, I don't know all the theory. Such guessing generally seems to work better when the clipping is not too extreme; that's why I said what I said about "not too bad".xstylus wrote:While there's varying severities of clipping, there is no way to "fix" clipped audio. If you've got a song that's clipped but "isn't too bad", the best thing to do is leave it alone. The only true way to fix clipped audio is to recapture the audio and pay closer attention to the levels.Though I know there are programs out there that can fix clipping if it's not too bad (Adobe Audition comes to mind; I don't know if it had this functionality when it was still Syntrillium Cool Edit Pro).
Clipping is exactly what it is. It results from levels being set so high that it hits against the peak, or "ceiling" as I call it. When the levels are cranked up, anything above the ceiling is chopped away like a lawnmower cutting a blade of tall grass.
Even if you bring the levels back down, that doesn't bring back the parts that were clipped off, and thus such parts of the song sound flat and distorted because some of the sound data that was there is gone.
When Adobe Audition attempts to fix a clipped section, it does NOT bring the levels back down at the same time (unless you tell it to); it tries to fill in the missing sections of waveform by adding samples ABOVE 0 dB FS. It leaves the waveform like this until you either normalize back down to below 0 dB FS or save (and if you save first, then of course you're just chopping off the peaks again, defeating the purpose of fixing the clips in the first place).
No, that is NOT normalization, that is <i>dynamic range compression.</i>milkmandan wrote:A great example of this is Linkin Park's - What I've Done song.
If you've ever heard it on the radio and then heard it from the CD player, you'll notice the radio version is normalized, as the intro piano is at the same volume as the chorus. In actuality the piano is quite soft. You lose that dynamic audio range.
Normalization scales everything evenly to fit a given peak sample value -- it either makes everything louder or makes everything softer. It is not for evening out levels between parts of a track.
If I can do AX work on my b-day so can she. 
