Cubase 11 ARA folder is gigantic from using Melodyne

Hi,
I can’t see that this has been discussed before, but please direct me if this is a known issue or work as designed.

I have a project with some lead and backing vocals on several tracks being comped using the track lanes function. After comping I touch up the vocals with Melodyne to adjust pitch and timing, on the same track without rendering to a new track, which works well enough. However, the ARA-folder becomes gigantic compared to the audio material it is editing.

The vocal takes total around 1.5GB in total and in general I have 5 takes per track over numerous tracks. Naturally, I can only use 1 take per track in total, so that takes the effective use of vocal tracks down to around 300mb (1500/5). The way I understand Melodyne, is that it makes a copy “seamlessly” through the ARA protocol, which I then edit using Melodyne in Cubase. So, I can accept that the ARA folder becomes as big as the audio I’m editing (300mb) plus some overhead, let’s be generous and make it an even 400mb. But the ARA-folder is 10GB! 25 times the size of what I would expect. Maybe I can accept that it copies all audio on all track lanes, which then, with the overhead would take it to 2GB (400mb x 5 lanes). That is still just 20% of the current folder size. So, why does it need to create a 10GB folder from 300mb material? (Also, why are the file names in the ARA-folder gibberish, so there is no way to work out what file belongs to what take? It makes the fault finding mission a lot harder.)

Is this a known issue or is it working as designed? If the latter what can I do to avoid these massive folders in future projects.

Oh, also, it seems like the very large ARA-folder is making the project very sluggish. It has no trouble actually playing the project but struggles with updating the screen and simple things like open a folder track or even stopping playback of a project. I can take a minute for the project to actually respond to other commands again after stopping playback, showing me a nice little spinning wheel instead of my mouse pointer.

Any good ideas on this subject?

1 Like

Hi,

I have tested with a simple project (only 1 Audio Event) to make sure I have it under the control. I cannot reproduce it here the way you described. If I have 3.6MB WAV file and I apply ARA (SpectraLayers One), I get 3.1MB WAV file in the ARA folder (I would expect 3.6MB). If I remove the ARA Extension and apply it again, I get another one 3.1MB WAV file. So now the folder is 6.2MB.

This is the very same approach Direct Offline Processing has. The reason to don’t delete the older files is for the Undo function. If I would Und 2 times to get the 1st ARA I applied, the 1st WAV file would be in use.

What I want to mention, the takes are actually just a virtual cuts. If you record 5 takes, you create 1 long WAV file (which is 5 times longer than the recorded time), which is virtually cut to the takes. So if I recorded 2.7MB file (which are actually 5 takes) and I applied ARA, ARA was been applied to all takes at once. The ARA WAV file was 3.5MB then (I have to say, I don’t understand, why is it 25% bigger here).

But in any case, the files are not much bigger. The folder is growing by repetitive Extension usage.

What is the location on Win 10 of this ARA folder?

/Magnus

Hi,

It’s the same for both platforms:

  • Your_Project_Folder/ARA

Thanks, Martin. But I am probably just daft: Where exactly is this folder located? Do you mean the actual folder I made for the project? If so, there is no ARA folder in it. So I am a bit confused…

Best,

Magnus

Hi,

When you create a new project, you define the Project_folder in the Steinberg Hub. This is the " Your_Project_Folder" I refer to. So the ARA subfolder is at the same level, as the Audio subfolder, Edits, Images, Track Pictures, etc.

This ARA subfolder appears once you apply any ARA extension for the first time in the project.

So if you have a very large ara folder it might be saving in your default project location and not this specific project. That means that ara folder will have all your projects ara files in it, so beware before deleting…
Just a thought, not a fact (yet)

Hi,

How did you come to this, please? I’m not aware, there would be difference based on the ARA folder size.

Thanks for clarifying, Martin. That’s what I thought. Strange then that I can’t find an ARA folder in the project folder…

Best,

Magnus

It is easy to check, if your ARA folder is in the project folder it is fine. If it is another folder maybe not so fine. Just check and report.
It has nothing to do with size per se, but with location. Depending on how you start a project files might go to the default project location or the project folder. I am not sure this applies to ARA too but it does to where your audio recordings go.

Hi,

So do I understand you right, the ARA folder is placed at the same level as the Audio folder? That’s correct.

Of course, if you create multiple CPR files at one folder, the Audio (and ARA) folder is shared for all these project. Therefore it’s highly recommended to create dedicated folder for every single project.

Hi,

I can understand that, even if I would think with the ARA-protocol it would be a bit smarter so that it could do the edits parametrically. At least it should understand to only create a copy of the take and not the whole wave-file. It’s not a very elegant solution. It seems more like a brute force solution.

So, I’ve run a small, controlled test to see what is actually going on. Do you think this file handling is reasonable?

My biggest question is why Melodyne needs to create a copy of the whole file and not just the take that is being edited. It’s impossible to use two overlapping parts of two takes anyway, unless I create a duplicate track. So, it should only need to copy the take and not the whole file. It’s not like it is preserving file names anyway.

The workflow I use is to record a number of takes, comp them to get a best composite take and only after that edit pitch and timing with Melodyne.
I might record 20 takes but maybe just use parts of 4-5 takes to create a composite. Why do then Melodyne need to copy all 20 takes 4-5 times? Shouldn’t it be enough to make copies of those 4-5 takes?


TEST
Empty project 44.1kHz, 16-bit wave, tempo 120bpm
Create 1 audio-track. Recorded some audio for 3 bars in 3.75 takes.



Select take 3 and use Melodyne as extension - no edits
Gibberish-name wav-file (4066kb) shows up in ARA folder almost 4 times larger than the original (3051kb/3.75takes) take (4066/(3051/3.75)-1=3.9975) that is used.
Make 2 edits (pitch + timing) - no increase in size of file in ARA-folder.
Select part of take 4 and add the Melodyne extension to this part of take 4. A new gibberish-name wav-file is created equal in size to the first ARA file. Now I have 2 ARA-files that are almost 8 times larger than the actual comped take.

Melodyne can’t know what part of the take I want to use, so it is probably best to create a new file of the whole take, but it is completely unnecessary to create a copy of the whole wave file, as the takes cannot overlap.

What if I have 20 takes in a file? Then I get ARA-files that are 20 times bigger than what they need to be + the 33% headroom that it needs for some reason (4066/3051-1=0.3327).

So, if I use a small part from each take, I can understand that it needs to make a copy of the whole take, but not 20 copies of all 20 takes. It makes no sense.

It seems like a very inefficient use of resources, both disk space and processing power to manage all these files in real time. The result is that you get an ARA folder that is gigantic in size, which slows down Cubase so that it almost grinds to a halt. The odd thing is that there is very little disk activity going on and yet Cubase can barely function in a fairly normal 75-track project, with a mix of audio, instrument, and midi tracks. We haven’t even started mixing yet, so not a lot of effects as inserts and/or aux.

I wouldn’t care so much if it was just disk space that gets eaten up, but when it effects performance, it becomes a big issue. I’m inclined to ditch Melodyne in favour of the build in equivalent, which I feel is not having these issues and is a bit more clever, even if it doesn’t sound as good and is not as intelligent for pitch and timing correction.

We’ve tried this on two different computers, a high spec Dell Precision 7530 workstation (see below) and a maxed out Azus self-build 18 core, 64gb RAM, etc.

Any clever ideas on:

  • Why this is necessary?
  • What the best workflow is given this wasteful use of audio files and storage?
  • is this a bug?
  • is this down to Cubase implementation of ARA or Melodyne’s ARA protocol?

I don’t know much about comping with lanes but I feel I’m doing the same thing with tracks vs lanes. I make several passes of the same phrase and then piece together what I need from the different tracks to make one usable take on one track. That way, Melodyne doesn’t have to decide on anything but that one track/line. I guess the difference with lanes is you can hide the clutter but I usually delete the unused files after I get what I need.
My take is that the problem with doing it with lanes is that Melodyne’s ARA protocol can’t differentiate multiple parts in different lanes on one track as well as processing a single track itself. Therefore it has to accumulate all the information (parts, lanes & tracks) to make sure it has all that you intended to work with. As you can see, this becomes very accumulative and expands the folder working area.
That must tie up a lot of buses going to and from and compromises Cubase’s intended pathways and ultimately locking it up. Therefore, it’s not a processor problem but makes it seem like even the most proficient computers are getting bogged down even when the processors are hardly working .
I haven’t tested this theory but it’s what I see from the outside looking in.

i think you may be right, but it’s not a very clever implementation. You may just as well treat the lanes in the same way, in fact even more exclusive, than tracks, as lanes cannot overlap, but tracks can. So, if you select 1 of 4 takes, it should only convert the part of the wave-file where that take is located. There is no need for it to convert the rest of the file, as takes cannot overlap on lanes, until I actually want to use a part of another take. So, it is creating 3 tracks of audio unnecessarily, from just selecting 1 take to correct in Melodyne.

Alternatively, it could convert the whole file with 4 takes, which is a bit less elegant, but then it should actually use that whole file when I select different takes to apply Melodyne to, as it has already converted that wave-file. But currently it doesn’t. It makes another copy of all 4 takes. So, if I select parts from all 4 takes it has converted wave-files for 16 tracks, 12 tracks of audio completely unnecessary, and you may argue it is actually 15 tracks of unnecessary audio.

When it is 4 takes it may not be so much of an issue, but when you have 20 takes, where you for example try out harmonies of 5 takes of each part in a 4-part harmony in one single recording (so also have some takes for doubles), which is not uncommon today in pop-vocal production, with the objective of duplicating these takes 7 times so I end up with 4 tracks with 2-3 takes on each. Melodyne/Cubase will create (assuming I use parts of 2 takes for each harmony part) 8x20x2 = 320 tracks of audio for a 4-part doubled harmony, where it should be able to make do with 8 tracks in total.

Is there something I’m missing here or is this actually how it is supposed to work? It would be great if someone from Steinberg could provide some insight to how this workflow should work.

I can’t remember that Cubse own built-in pitch-correction feature, VariAudio, works like this. As it is half-decent, I may be better off going back to using this and leaving the €700 Melodyne Studio plugin in the drawer (or sell it), until it is really required.

Variaudio is great. I use it for vocals (and prefer it) over Melodyne although I only Melodyne Studio 4. I use Melodyne for polyphonic stuff and fixing timing mostly.

I bought Melodyne before I had a full version of Cubase so I have never used Variaudio. I’ve had it for 4 versions now and maybe should check it out for its strong points. Just got Auto-Tune 9 in a sale bin for 5 dollars. Between the 3 there should be no problems.

Donald, I feel you on that issue with Melodyne. You would know better than I because I hadn’t used the lanes so everything is pretty solid on this end. I, like you, am waiting for Steinberg to chime in on this one though.

So, how do you actually comp vocal takes if you just use tracks? I guess you have to do a lot of mute/solo-action to first audition the various takes. And then copy and paste the parts you want to use to a new track. Is that how you’re doing it?
It seems like a rather cumbersome workflow compared to lanes recording and the “comp” tool, as described in this short clip. Or am I missing something?

The ARA/Melodyne implementation is certainly more reliable than it used to be, but @ dlinderyd you’re spot on with how it works and it is indeed very wasteful of disk space. I expect this is one more iteration in many so I’m sure they’ll improve the situation. My suggestion is, as you say, to re-use the ARA wav for all events which are sharing the original wav file. That’s the way it works when you use VariAudio.

Off topic - I tested VA vs Melodyne recently and found I preferred the sound of Melodyne but it was so close. For me it was a tonal difference and I preferred Melodyne’s tone, it sounded a little bit warmer.

Mike.

Not sure what’s the final finding now about the amount of files inside the ARA subfolder (only created if any ARA extension was used), but what I think is an issue, that all files stay in the ARA folder, even if I do not save my project and on close I confirmed to delete any created audio. It may then happen, if you work in a same folder but you frequently dump projects the ARA folder can get very loaded. To then find out which file to keep and which can be delete can be a challenge.