I would like to say what I think these mean, but be warned that I am just guessing.
Quote:
First off, what do these percentages entail? For, say, "stop declipping if peaks did not increase above", what does a value of "100%" even refer to?
This one is talking about a found flat peak. All samples in a music file can be defined as a double floating point value from -1.0 to 1.0. (I would use pictures but have nothing good to host them on) A flat peak would look like this if we just look at the samples of one channel: 0.3, 0.4, 0.5, 0.5, 0.5, 0.4, 0.3 (Ultra simplified for illustration purposes.). The clipped section is the part where all the 0.5s are. That 0.5 value is the 100% mentioned in the setting. The reconstructed peak should probably look like: 0.3, 0.4, 0.5, 0.6, 0.5, 0.4, 0.3, but if the declipping process comes out with something that's the same as before (0.3, 0.4, 0.5, 0.5, 0.5, 0.4, 0.3), that section should be ignored. This setting will let it accept lower value results (0.3, 0.4, 0.5, 0.4, 0.5, 0.4, 0.3) by entering a lower percentage, or force it to only keep higher results with a higher percentage.
Quote:
"maximum deviation from straight line", I do not understand what a value of, say, "50%" would encompass.
I think this one is talking about little bumps in clipped lines. Many times in recording, the clipping will be applied on just one input (Like drums because they're pretty loud compared to other instruments), but after mixing and other processing, the final result has little deviations from the straight clipped line. I'm also not totally sure what it would encompass, but my guess is that having a greater percentage allows longer flat sections with little bumps or spikes to be treated as a single clipped peak.
Quote:
what do the terms "probably" and "possibly" in the "Sample probably/possibly clipped if above" sections mean?
This percentage is talking about the maximum dynamic range of a sample (-1.0 and 1.0), but can be adjusted with the dynamically increase level option. A really good recording will not have any samples above 0.95. (I just looked at one of the best recording that I have, and the highest sample was 0.9432.) That will not be the case with about 98% of the music StereoTool will be processing. Most recordings these days have multiple samples that are right on the 1.0 line, meaning they are almost guaranteed to be clipped. These two sections "probably" and "possibly" are for the case where you have the input volume turned down to reasonable levels. This means that the highest sample fed to StereoTool is only ever like 0.7 (Even though in the actual file, that 0.7 is really 1.0). I recommend using the dynamically increase level options on those because that way it doesn't matter what the input volume is, it will still be able to figure out what should be looked at for clipped sections.
Quote:
Say I want to declip the single "Take Me Out" by Franz Ferdinand, which is highly dynamically compressed.
If you zoom in, you can see a huge portion of the signal makes a pretty flat line as it goes up and down.
How would I make sure to get this portion of the signal?
This is what the tilt section is for. You can specify the tilt degrees in the Input section to make sure that that section gets considered in the declipping process.
Quote:
apparently the "MP3 dirty area" in the "Input" Section is active regardless if the "Input can be compressed" section is off.
It might add a tiny bit to the processing requirements on your CPU, but I have not found a song yet that sounded worse having it enabled. I usually have that value increased to like 5. I've been able to find settings that do not change the sound of good recordings, and considerably improve clipped ones.
Sorry I don't know anything about the VST plugin. I mainly use the command line version and then the standalone version for optimizing settings.