IMO it is not clear if the problem is the realtime algorithm that does the pitch shifting, or if the problem lies in the analysis part (which is not done in realtime as you know).
The Variaudio algo is licensed from zplane. Last documented (and audible) update in Variaudio audio quality was in Cubase 7.05, see here. So I assume the algo did not change since 2011, or at least not in an audible way. This doesn’t mean anything bad - why fix something that simply works?
Variaudio 2.0 and 3.0 just added much better handling and integration inside of Cubendo, but I bet the audio algo stayed the same since 2011.
Last updated docs by zplane are from 2021 and available here, and according to the docs, v3 didn’t change their SOLOIST/ TUNE algo (which probably is the one you can hear using Variaudio). The docs from v2 are here, so between 2011 (release of Cubase 7.05) and 2021, simply not much seems to have changed about the algorithm.
ProTools seems to use elastique v3 since last year. I did not find any information about Steinberg using v3 yet, so probably Cubendo 13 still use Elastique Pro v2.
zplane released a new product named INTONE lately, which is based on the soloist algo, but also includes the analysis part in realtime - now this would be interesting if it found its way into Cubase. Record in audio editor, pitch correct in realtime, and later edit the corrections if you like. Best of all worlds.
Lots of question marks. If the analysis part of Variaudio is the key to solve this, one would have to know if this is a Steinberg creation, or if it’s also part of elastique. Or simply ask zplane if noone of Steinberg gives any details.
Get yourself the demo app from this site (scroll down) and select soloist mode. Load an audio file and listen to stretch and pitch. That’s their 2024 algo. To my ears, it sounds exactly like Variaudio does today.
If it simply worked I wouldn’t be posting here!
Thanks for all the info but I have sneaky suspicion that neither reading all that documentation or contacting Zplane is going to get me any closer to having a working VariAudio!
I ran some of the same tests in v5 and got the exact same results so whatever they did in 7 obviously didn’t affect this.
The biggest question to me is not the analysis but why they can get something to work in real-time but not be able to render it.
Are Zplane also responsible for the algorithms in fine-tuning and pitch-correct, which both exhibit the same issue, ie. they don’t render correctly?
I don’t know. If the problem exists when offline rendering, but not during realtime playback, it could be any of the components, even the host software’s rendering process. What happens if you create your mixdown with ‘realtime export’ switch selected?
Exactly - that’s why I’m wondering how many other features may suffer from the same problem. No, real-time export doesn’t work. I only tried that after Cherry Audio found that for me as a solution to automation on their Voltage Modular synth not rendering properly in Cubase. I got very excited that might be the solution to this too but no such luck
Didn’t SB include a light version of Melodyne at some point? Or am I misremembering? I have a light version and pretty sure it came with CBPro 12. I wonder if this is why they did that?
Sorry, I didn’t have any time during the last few days. I wanted to download it today but it seems the file is no longer available. Can you re-upload it?
Ok, I think I found something, which might be the cause for your complaint.
To me it seems that the algorithm has some sort of “grid” when it comes to frequencies and the grid becomes more apparent the higher the frequency is.
I looked at your A4 (and also mine), which is around 880Hz.
In VariAudio you can set the “Pitch Snap Mode” to Off
and then drag the note up and down, changing the frequency by a small amount only. It works best when vertically zoomed in as then you can make finer steps.
Using the Tuner or MTuner, that @Held mentioned, and of course my ears you can actually see and hear the steps pretty well.
E.g. an A4 +38% in VariAudio is shown in the plugins with 892Hz (A4 +23).
An A4 +39% becomes 911Hz (A#4 -41).
The next frequency change happens when dragging the note up to A#-25% (plugins: 930Hz, A#4 -5) and so on.
From A4+39% to A#-26% the frequency of the note remains at 911Hz.
So the frequency changes in VariAudio are not fluent but they are happening in steps. The higher the frequency, the more Hertz between the steps.
That’s pretty appalling! I don’t know that I’d say it’s the cause rather than yet another symptom and yet another sign of how unprofessional it is. If I’m understanding what you’re saying it would mean that much of what you see on screen is just for show only and shouldn’t be taken seriously but it still doesn’t explain why something can sound perfect in VariAudio in a project but out-of-tune when it’s rendered
You are entitled to your own opinion. While I also find the current algorithm not awesome I’d say it seems to be working better in the range of the human voice. For this it seems to be good enough. Still, it wouldn’t be bad if Steinberg were to do research on the algorithm.
This is the part that I examined first. You remember me mentioning null tests a lot? You mostly waved that away as mumbo jumbo, which I honestly did not appreciate at all. I have found no instance where the rendered version differed from the real-time output. So this is something that you have exclusive. Feel free to provide a project with audio in it, which showcases your claim.
My opinion is that there’s not much point in having analysis if it’s not analyzing correctly; there’s not much point in being able to edit something if you can’t then render it, and from what you were demonstrating, there doesn’t seem much point in having a snap on/off if it’s still moving in random jumps rather than smoothly.
Yes, I certainly do, and I’m sorry you feel that way but I didn’t say they were mumbo jumbo, I said that in this situation they were mostly a bit of a red herring, ie. a distraction. I have used null tests plenty of times, when they were relevant. In this context that was only to see whether two files which should have been quite different were identical or merely very similar. They showed the files were identical except the ‘edited’ files had a 2ms fade at either end, which is probably what you were asking about earlier, and which showed that some processing had occurred but certainly no pitch correction. If two files are 30-40 cents apart I don’t need anything but my ears to tell me they are different. I do not need a null test, in fact it would be utterly pointless.
I already have provided you with a project with audio which showcases the issue, twice in fact. It took me a couple of hours or so to put it together and I did so out of politeness and respect for the fact that you were engaging, not because I felt it would be of any use. If it wasn’t what you wanted/needed then perhaps it would have been a good idea to answer my multiple requests to be specific about what you wanted.
I did not make the original post because I needed people to confirm if I was right or in the hope that someone might have a solution. I know there is a problem, Steinberg have admitted to it and I’m sure if there was a workaround they would have told me.
I made this post to warn people they were working with software which they shouldn’t rely on to be doing what they think it is, and that in certain circumstances it may make things worse rather than better, so be careful. And that warning goes out especially to anyone who may feel less than confident in their own pitch sensitivity and think they might be better off snapping everything to 0 cents to be safe. You are NOT safe.
What you or anyone else does with that information, or whether you believe it or not is entirely up to you.
Yes, thank you. That was helpful for my tests.
You raised two points:
1.) The initial analysis of VariAudio is faulty and subsequentially any moved note may or may not be on point.
I think I documented in what way that can be heard and reproduced by everyone.
2.) The other point was that real-time playback sounds different than the rendered file. This is also the title of this topic.
And for this I found no evidence, also not in your project. Except for the fade, which I will investigate a bit further, but I am not sure that this is correlated to VariAudio.
I have a rather technical background and like to differentiate things where they differ. I am also trying to break the information down into, if you will, their elemental parts. Maybe it’ll give a Steinberg developer a point to think about when looking into this issue, which they previously might have missed.
There are lots of problems with audio warp as well and it may well be related.
What you see on the screen when aligning transients to the grid, is not what plays back or renders.
Here’s one thread -
Shifting in both time and pitch seems to be at best just approximate, and we’ve known this for a number of years now. I’m sure it can be improved, but whether we’ll ever see it in the current iteration of Cubendo is now doubtful.
Firstly I owe you a big apology both for being a bit snappy and dismissive in my last reply but also for not checking what I sent you properly. I did it by numbers, without really listening to anything and although I sent you notes on what pitches were being shown on my system I couldn’t test the parts that were still active in VariAudio outside Cubase. I’ve now installed the MTuner which was useful, although it seemed to think both my in-tune and out-of tune B4s were slightly sharp E4s - very odd! But if I’d bothered to listen I would have seen what was really going on just with me ears.
What’s actually there just makes matters worse. You’re right - the renders didn’t change, not on this test. Instead what I have is, for the worst example, one B4 which reads as having an original pitch of -6% (but should correctly read as 0%) and current pitch 0%, and one which has been rendered from that which reads as 33% sharp, whereas in reality they are both 39% sharp!
So no, those tests don’t prove the rendering thing but although I don’t have anything on hand to prove it to you (the testing I was doing with vocals was two years ago) I can assure you that I definitely have experienced this too. This whole thing kicked off because of one high vocal note from a part I was editing in one project and importing into another - something I don’t normally do so would usually only ever hear a rendered VariAudio part in a final mix. Every time I pulled it into the main mix it was sounding very sharp so I’d go back and have a look - could never find anything sounding wrong but would mess around for a while and try again but the same thing always happened - fine in the edit, the rendered one sharp in the main project. The original note was a little out so I needed to do something but in the end the only way I got round it was by rendering that one note with a stronger setting on Autotune and leaving the VariAudio off. So sorry, no, I can’t give any supporting files at the mo but take my word for it it is also an issue. I think @vinylizor 's post also backs this up. There seem to be so many things that are wrong with it that might randomly come to the fore on any particular part that it’s not always easy to keep track of which particular wheel has come off the rails.
