It’s an interesting discussion. Personally, with my very amateur involvement with the tools, I’ve found that for me it’s easier to tune voices than instruments.
And, for better or worse, I find that leaving things as they are is sometimes better, IF the performance is self-affirming. There are some notes, sharp ones, that are exactly what should happen, because everything else at this point dictates it. It’s more sentimental than proper. But other times, a pitch might wander up or down unintentionally. These are what I try to correct, but easily sliding the event up or down does not always achieve the purpose. The line of the pitch should also be taken into account, and the nature of the instrument to tune.
Guitar was a great example. Does anyone bother to tune each note? Most are off. Flutes, trumpets, trombones, saxes, everything has its own “blueprint of wrongness”. Each and every note has its quirks. Of course the player strives for good intonation, especially in ensembles, but even then, it’s not a matter of being in the green of the tuner.
Improvements on the side of the algorithm and analysis are welcome, of course. But I agree that the tool is more than good enough for monophonic sources.
(Imagine if we had Polyphonic Variaudio for free in Cubase. A scandal!)
If you add amplitude it’s quite literally the definition of music!
Like it or not, every tool you have to express that emotion is built on maths, whether it’s the VariAudio or the voice itself, whether it’s numbers on a screen or the maths inherent in nature.
It’s a question of semantics really - for some people if someone sells you a car and doesn’t actually say everything works fine but merely doesn’t tell you the steering’s off and hopes you don’t notice they’re not conning you, it’s just your bad luck.
Steinberg know there are issues, they don’t tell you that but advertise everything as being full functional and professional. I do know that much.
You’ll have to excuse my ignorance but I don’t get the Yamaha reference - did they create the algorithms?
But if you want to be specific what I actually said was that VariAudio was a con - I’m happy personally for you to interpret that as me accusing Steinberg as perpetrating a con but what I also meant was that VariAudio cons you into believing things are in tune when they are not.
I’m familiar with the idea that some software is better than others - and more expensive, and I’ve never complained to a company just because something’s a bit crap, and I’ve bought plenty of crap software over the years, but if it’s not doing the job accurately don’t tell me it is, either in the software or in the sales blurb.
If that’s referring to me you’re way wide of the mark! But seeing as you posted an RIP for Bert Weedon I’m not going to risk betting I’ve been making music since before you were born, like I would with most of the contributors!
I’m totally onboard with the idea that human impreciseness in both pitch and rhythm are an integral part of most music (but not all). And that choirs and string sections wouldn’t sound the same, or as good, without pitch variation, nor would detuned synths be a thing (or chorus pedals).
I understand that equal temperament is a compromise - that mathematically a G sharp ought to be about 24 cents sharper than an A flat, etc.
And I certainly understand that we have a tolerance within which neither us or our audience is likely to hear something as wrong.
But I think some of you are missing the point that I didn’t start looking at this because of some sort of academic interest but because certain things were sounding BAD. Only noticably so on rare occasions but even once is a problem.
Whether a vocal note is in or out of tune needs to be down to me or the vocalist, not at the random will of software.
And of course there is a limit - a note that has been pushed to 40% sharp is always likely to sound bad, not like it’s been ‘sung with feeling’!
I’m also well aware that my sensitivity to tuning might be a lot better than most people but still less than plenty of people who might hear my music, and because of that it would be nice to have a tool that I could turn to to correct something if I think it might be worth it without worrying that it might end up worse than it started, either before or after rendering.
That’s why I loved VariAudio when I first used it - because it was able to do even large shifts (I use it for re-pitching, not just correction) and still keep it sounding natural. I was absolutely raving about it at the start - it was a game-changer for me, and I used it everywhere. And don’t get me wrong, I still use it a lot now but I would no longer recommend it to anyone else without warning them to be extremely cautious.
If VariAudio could always be relied upon to get within about 10 cents of the desired note (and stay there) I’d be completely happy still (probably!) - because I doubt I would have ever noticed - but it couldn’t, and I did!
What uncertainty lies with the algorithm is less than 10 cents. You can easily check this out yourself by spitting out sines, from 438 to 440 with a 0.1 step each. You will see that 438 is detected as -8%ish, 438.1 -7ish, but from 438.4 to 440 it detects A3 +1% (!)
Disaster! Ok. But if you check what 438 to 440 is, it’s less than 9 cents, and 438.4 to 440 (where the algorithm returns the same pitch) is some 6.5 cents. So, I’d say basic detection is good enough.
But why do we leave the Pitch Curve out of this discussion?
To be honest, particularly as it doesn’t analyze correctly, it’s not always easy to tell what’s happening when, and what I have realized, largely thanks to @Johnny_Moneto, is that at least some of the errors I thought were happening after the rendering were actually happening before. I can tell you that when I first noticed the problem two years ago it was because of one note that seemed to be sounding fine in the edit but was always rendering very sharp, but most of the testing I did subsequently was focussed on the numbers, and although I knew by then it wasn’t analyzing correctly I didn’t realize till now how random that was - I’d come to the conclusion it was reading everything about 6 cents sharper than it should, not that it varied wildly.
There are other posts on this forum that seem to have come to the same conclusion - that the faults are happening during render - but at the moment I can’t offer any evidence to support that, and because I now realize that there are a lot worse problems pre-render than I had thought, thanks to the responses here, and further tests I’ve done, I will admit I’m no longer 100% sure about the conclusions I came to re. the rendering. However that simply means I may be wrong about the source of the problem but the size of the problem actually appears to be worse.
Not at all - example: B4 sine wave - reads in VariAudio as -6% (which it’s not) - snap it to 0 and it now plays back and renders at 39 cents sharp although the render still reads as 6 cents less than that at +33%. 39 cents is way too much!
I use the Straighten Pitch Curve tool sometimes but I don’t remember noticing any specific issues there
Is it the analysis side that you guys think is the problem then? I can repeat @ggmanestraki test as per the screenshots and get similar results.
But if i leave the straighten curve setting on and move note up or down and then return it to it’s starting position it seems to pop back to 0 cents for me.
Seems like there’s some little quirks that exist, most likely more noticeable ‘by ear’ on a clean sine than more harmonic content.
For @alexis above, I setup 3 tracks, I used something other than retrologue to rule it out:
Track 1 : Arturia Mini Instrument track playing sustained note on Osc1.
Track 2. Bounced audio from track 1.
Track 3. Track 2 audio, as a new version (So it’s own audio file).
Turned the volume for each track down, added a tuner for each (Stacked so i could see them all at once), played the project on a loop.
Then, made adjustments to Track 3 via variaudio and the third tuner went out of tune to the other two. Big audio difference too when compared to the original tracks.
The track changes to solo mode too when using variaudio, so perhaps the process of that engaging is a part of this puzzle too?
My test is the following, and anyone’s free to tell me if it’s an alright test, or crap.
1 audio channel with a test generator. Goes into a group.
1 audio channel the input of which is the group, so that we can record what comes out of the test generator, and a Tuner insert, so that we can check the frequency during playback.
I manually type in the frequencies, the output of the test generator is -18dBFS and a sine wave. I calculate these frequencies assuming perfect octaves, and A = 440 Hz, and each semitone is 440 * 12th root of 2. This agrees with online tables of pitches, so no need to use a calculator, just look up the frequency of each pitch on the internet.
Then record what frequencies or pitches we agree upon, and share our results?
(I also thought of making the test generator “sing” a specific tune with automation if possible, but quickly dismissed the idea as too much work.
This is what peaked my curiosity too. Is the solo algorithm involved in the analysis, or not? The variaudio icon goes out when no corrections have been made to the detected pitch, but as soon as even the most slight adjustment is made, there can be (depending on the situation of course, that’s what we’re trying to find out) a spectacular effect on the original pitch.
I also wanted to avoid VSTis altogether, because I don’t know exactly what’s going on under the hood. Well, I don’t know what’s going under the hood of Test Generator either, but it seems to me a safer source to use (for sines).
That’s a good question, and I’ve also wondered about that. My guess is that when a pitch is obviously out of tune, like 10 cents or more, it gets correctly analyzed as such, and attempts to correct it work fine because there is more “room” to move the pitch towards the “right” frequency. Just a guess, might well be nonsense.
This has already been reported to Steinberg directly. They have decided that “it won’t be changed/fixed anytime”. This post was so that users might also be aware that it does not work properly.
