VOLOVIA wrote:The 'granulation issue' (it sounds like a BBT episode) is recurrent.
Ok, I had a quick look/listen to your files. From the differences between your two files, I can clearly hear what you are referring to in your source files. It does indeed sound pretty bad.
VOLOVIA wrote:I have developed a theory: when the 'raspy' voice, i.e., kind of pink noise is present and Flex is called on to pitch a note around it, if it is pitched up the granulation occurs (by stretching the 'noise', even if it is not included in the 'box'), while if the note is sharp (thus lowering the pitch), clicks are generated (because it time compresses the sample). This is the example that I include below.
I'm not sure I agree with your theory, but I don't really know the exact DSP that's going on under the hood - it's certainly not quite that simple as there are both format shifting and unpitched signals going on. Let's skip over our speculations for now.
Anyway, I loaded both your files into Logic. Firstly - you've got a great sounding voice! It's only a short sample, but there's a weird honky harmonic thing going on there in the middle (nothing to do with noise) and that probably isn't helping the flex pitch algorithm.
In any case, turning flex pitch on on the recording, and playing with various settings, pitching things around and so on, and I could not duplicate your "granular" effect that you display in your "after" file. With flex pitch on, I would say the audio was marginally affected (which is fairly normal) but not overly so, but regardless, with both zero pitch change, or 100% change to absolute pitch, with or without formant processing, I do not get that breaking up "granular" thing anywhere near what you display in your after file. It *is* sort of there on the source file - vocals can do some weird things harmonically sometimes, and as I said that's just a factor of the recording (for this particular sample, at least), but I can't get anywhere near to your "after" sample whatever I do.
If you format shift the vocal up quite a bit as an excercise you can get a different perspective of what's going on harmonically in that section too - this will sound artificial of course, but it's interesting nonetheless.
(Of course, I don't know what you did to get the "after" sample without the actual project, so it maybe that you've done some other processing or stretching which has contributed to this. Certainly if you get that output just by turning flex pitch "on" on that vocal, then I cannot replicate that here.)
I can do some some samples if you don't believe me. If you want to investigate further, it's more helpful if you could make a simple project example with that file and your flex pitch edits that's producing this effect on your system, and I can then see exactly what you're doing, and see whether it produces the same effect here, and that may give some clues as to the weirdness going on.
I was going to compare the results with Melodyne on the same phrase, but given Logic isn't giving me the artifacts you are getting I'll skip that for now.
What version of Logic are you using? I also notice the files are 16-bit - I don't know if you've bit reduced these down just for demonstration purposes, but I strongly recommend recording at 24-bit.
VOLOVIA wrote:It is extremely annoying and the only way to overcome it is to cut the word out and pitch the rest.
There are indeed times where DSP on acoustic material can result in various artifacts (not just in Logic's flex pitch, I mean in general). I'm particularly sensitive to this - try seamlessly looping thousands of difficult to loop samples - it's really easy to confuse the brain. I have certainly found that while I don't tend to use Flexpitch (or flextime for the same reasons) that much, when I have done I've found the results... variable. Sometimes, for some audio, and with what needs to be done it works fine, others, not so much. There is enough doubt for me that the results will be ok that I tend to use different tools for these tasks that have a (for me) more predictable outcome.
VOLOVIA wrote:Unfortunately, if the word contains the 'husky' note, there is nothing one can do but to sing it again (maybe a blessing in disguise).
Or try a different tool or approach - there are a variety of ways you can repitch things. There was a time before Melodyne etc :)
VOLOVIA wrote:Also, the most puzzling issue is that you cannot undo the created artefact: by 'undo' the pitch returns to its original state but the noise, clicks, remain. You have at this point go back to the original wav.
Did you read what I wrote above regarding flex pitch resetting? Turning flex pitch *off* complete should play the audio unprocessed (ie normally) so you can always split the vocal at the word or phrase causing problems and keep flex pitch off for those - from what you wrote above it sounds like you are doing that already. Sometimes you need a composite approach on material that's proving tricky to process.
It's certainly a weird one, but I can't replicate it here yet, I'm afraid.