I have a song I want to lower by 5 semitones. I have done that in Cubase many times. I thought I ought to be able to do it directly in SL.
I first unmixed the song to drums & other (because I don’t want to shift the drum pitches, obviously)
I have the “Other” layer selected. The song is originally in Eb. I want to go down to Bb. I do a select all with the “Other” layer selected and soloed.
I long-click on the transform tool, to put in in “transform” and not “transform selection”. At that point, I see the parameters right under the menu bar. I pull down the note icon rather than the %, and I set the pitch box to -5.
At that point, I see the selection rectangle drop down from the top and the “Transform” dialog shows the function being performed. The audio sounds like garbage, but it is still in Eb. Time shifting is set to 0 and stretching is set to 100%
This can’t possibly be right. Any idea what I am doing wrong? It is as if it is shifting the overtones, but the fundamental pitches are remaining as they were in Eb.
===============
After some more experimenting, it seems like the clip was being duplicated – I could hear both the Bb and Eb material. That’s really strange because it sure seems like it duplicates when I have the “Duplicate” button off, and does not duplicate when I have the button on – seems backwards. It doesn’t much matter because the audio is really distorted. I think I had the “preserve formant” on when the sound was so distorted. With “preserve formant” off, it is better but lower quality than I expected.
The input was an MP3 that was generated by Suno. After wandering in the wilderness for about 90 minutes, the final result wasn’t terrible, but there was a serious loss in quality – more than I expected. For my purposes, it was very usable because I was just trying different keys to throw at a singer.
The original Suno AI track was higher than any singer I would be able to use. It sounded very natural, but any singer I know would have really strained to sing that high. These are the kind of dilemmas we will face with AI. AI can show all sorts of things that seem natural but really aren’t possible in the human world. (In this case, there are some screaming rock singers who could pull this off, but it would be 1 in 100.)
Anyway, I think my problem was an unfamiliarity with the UI for the transform function. The color scheme made it seem like the “duplicate” button was reversed. And the worst of the distortion came from having the “formant” switch on.
I think I will go back and try unmixing drums, vocal, and other and process as:
drums: nothing
other: just the pitch shift
vocal: pitch shift + preserve formant.
I bet that will be a happier result. I hope that is effective because, as an arranger, I often go through these negotiations with singers, and I want them to hear the best version of the proposed key before I ever do the first note of the arrangement. In this case, the original was in Eb. Bb was in a comfortable range for the singer, but the horns were not going to sit well in that key without a who different approach to the arrangement. But C could work for the horns. Turns out the singer things C will work for him, so I am good.
And I think I learned enough through this process to be able to do these transpositions entirely within SL, which is faster than using a combination of SL and Cubase. My first try was to do it all in Cubase with the new Cubase 15 unmix. But the quality was lower that way.
Here is the result after I transposed to C entirely in SL:
Notice when the singer comes in there are heavy artifacts. But more curiously, listen to the horns at the beginning. The brass notes are major in the original, maybe a little out of tune. But they are warped in the transpose function to where they now sound minor.
And I want to reiterate that I am not complaining, simply observing the state of the art. For my purposes, these results are good enough. But they really are quite degraded, which illustrates the limits today, and what we can hope to improve in the future.
hmmm, I’m messing with this a bit this morning…I’m just applying to an already unmixed backing vocal layer…it’s not pretty, for sure…I probably need to try on some more pure sounds…just figuring out how to get pitching to work with Transform tool is tricky enough…we really could use better guidance…I’d say user can “get an idea”, but it’s no Eventide harmonizer…I don’t do this type of editing very much…I think the KiloHearts vst might be better at this?
Even after doing this a bunch of times, I still find the UI really confusing. The “duplicate” thing seems to be an on/off button, but I can’t figure out which is which. But the “preserve formant” also seems to be an on/off thing, but it is a checkbox, which is more obvious. Why are they different? Really confusing.
And I have trouble entering text into the box where you can type the number of semitones. It doesn’t take all my keystrokes, and then I never know when the effect is actually processed.
I am glad Steinberg is including this feature, because it is the kind of thing a person would want to do with layers. But, as currently implemented, it is clunky and produces dubious results. I hope this area is able to improve in future releases.
I have never tried any of their stuff. My goal was to have a quick, easy way to give a realistic pitch change so a singer could make a good judgment about keys. In my experience, when I have given singers a poor, noisy example, they have sometimescome back later (after I have spent many hours carefully voicing the arrangement) with “You know, that key really doesn’t work for me”.
The two biggest challenges to make it realistic, I think, are:
Keep the drums at their original sound/pitch
Give the vocal track reasonable formants so it still sounds pretty natural (not Barry White or Alvin the Chipmunk)
I can do all of that acceptably if I use SL to unmix (that is better than the new Cubase unmix) and then use Cubase to pitch shift the “other” (that’s better than SL), and use Melodyne to shift the vocal (That gives full control over formants and has few artifacts). That is a really good result, still has a few artifacts, but really pretty nice. But that’s 3 different tools. In theory, it is all right there in SL, but it just doesn’t work well enough.
I just don’t do much music related processing with SLP; a bit of unmixing. Especially vocal removal and attenuating drums when they are too loud (always).
What I mostly use SLP for is NR for on location recordings for which I find SLP a jaw-dropping tool; so far beyond anything else I have ever tried. These NR tasks I speak of seldom happen quickly, but the results are amazing to me. I don’t do much else…I run a lot of Unmix Noisy Speech and then manually complete the editing. Yes, Unmix Noisy Speech is not a perfect unmix process, but an fantastic starting point.
I agree. I do videos fairly regularly to promote concerts for an orchestra I play in. We rehearse in a high school. And there is one location I like to use because it has a very nice background. But it is in a rotunda that has some strange acoustics. And there are also some very noisy vending machines nearby, and two intersecting halls where custodians are often using noisy equipment.
In the past I had to use a shotgun mic, unplug the vending machines and hope the custodians weren’t working the same time I was. With the newer versions of SL, the NR for speech is great. I still unplug the vending machines, but if there are other extraneous noises, it really doesn’t impact my recording. Kudos to the SL gang for this.
