At first sight it’s a marvellous tool – I fell in love with it the first time I used it – beautifully smooth and natural sounding edits, with no effort. But eventually I started noticing that every now and then, no matter how much I tweaked it, some notes sounded clearly out of tune when I rendered them, either in a project or as a mix. It was only very occasionally I would notice it, but my pitch sensitivity is good but not perfect, and more to the point this was happening even after the edited vocals were going through a subtle bit of Autotune afterwards, meaning they had to be considerably off.
It gets off to a bad start – it doesn’t analyze the pitch correctly. Render a sine wave from Retrologue at C4, E4, G4, A4 and B4 and VariAudio will tell you you have C4 +6%, E4 -3%, G4 +8%, A4 +4% and B -5%. Analyze the same files in Wavelab and it will tell you you have C4, E4, G4, A4 and B4.
The fun comes if you snap them each to 0% and try rendering (or flattening) them. The C, G and A all ignore the edit and revert to their original pitches but the rendered E and the B are now 26 and 40 cents sharp respectively (in Wavelab) – considerably worse than useless.
If instead you use it on something that actually is out-of-tune you might be lucky enough that it improves it. Do the same exercise with an A4 +50% and you end up with a note that is only 15 cents sharp (or 9% when analyzed by VariAudio) – hurrah! – but repeating the process on the rendered file won’t get you any closer.
I also just now tried something I hadn’t tested before – fine-tuned my 15 cents sharp A down by 15 cents, rendered it and guess what – it’s still 15 cents sharp!
This is how well it deals with something as simple as a sine wave – it doesn’t seem to be any worse with something like vocals but it’s just as unreliable. At the time I discovered this issue I found one vocal note that was +0% (in VariAudio) before the edit and +36% after.
None of the different ways of rendering make any difference. The only way to guarantee you get what is on the screen is to send it out of your interface and record the audio back in. Then it works perfectly.
The whole thing is a joke – or, as I said at the start, a con.
Most of us don’t have the sort of hearing that will necessarily notice something being just a tiny bit out of tune. That doesn’t mean it wouldn’t sound better if it wasn’t. And I think the vast majority of us are going to assume that if, just to be on the safe side, we’ve gone through an entire vocal line and snapped every note to the right pitch, and everything sounds fine, even if we can’t hear the difference, that we’re going to end up with what we see on screen.
I wish!
Depending on what you’re editing you seem to have a random chance that the rendered result will be 1, at least heading in the right direction, 2, exactly the same as before the edit, or 3, more out of tune then when you started.
I have tested this on different machines, and in versions from 5 to 12; if you put the same thing in you get exactly the same result, so the algorithms obviously haven’t changed in 15 years.
The thing that astonishes me is that you really would have thought that creating the algorithms to do the job in the first place was the hard bit – and I’m in absolute awe of the people who did the maths for that – and simply rendering it to a file would be child’s play after that. Apparently not, but if you can’t then what’s the point?
I first contacted Steinberg about this in May ‘22. It took 6 weeks before ‘tech support’ finally deigned to open a bug report with the developers. By that time I’d replied to multiple emails, done everything they asked plus a lot more testing and supplied plenty of evidence.
After another 3 months, and more prompting, I got an email saying ‘it seems they have not found any solution yet’, which at least was the first admission I had had that they accepted there was a problem.
That was the last I heard – I had no answer to my reply, which asked, amongst other things, whether working on the issue was a new thing or whether they had been trying to fix it over the previous 13 years (every time I had asked whether they already knew of the problem they avoided the question but I cannot believe I was the first to report it), and whether they had any intention of being open about the issue and/or stopping advertising VariAudio as a professional (functioning!) feature.
A month ago I noticed Cubase 13 was out, with VariAudio 3, so I wrote to ask whether they had now fixed the problem. This was what I got in reply - ‘Unfortunately, this report was classified as “out of scope” and it means it won’t be changed/fixed anytime’.
So there you have it – it doesn’t work, they know it doesn’t work and they have done for at least 2+ years but probably a lot longer, but my guess is that few enough customers have noticed and reported it that they can’t be bothered to try to fix it and have chosen instead to carry on misleading people and pretending it works.
And I do wonder whether that’s just the tip of the iceberg. Along with all the other things about Cubase that seem to indicate that Steinberg are a company that really can’t be bothered to make an effort, today’s discovery that standard pitch-shifting also doesn’t render properly just makes matters worse, and I have had issues with at least one third party vsti that works fine in situ but doesn’t render properly, so how much else is working fine in the program but not coming out the same way in the mix?
Be careful, and don’t take anything for granted with Cubase – that’s my advice.