Hello All, I am busy with the new Cantai plug inn. And it brings really something new to the table. But I am experiencing performance problems, especially when I am changing my score and work on the composition.
I have a M1 Mac Studio (M1 max processor) newest software (MacOS, Dorico and Cantai) and a Antelope Zentour (the older model)
Rendering takes quite a long time and the tax on the system is big. I only can have one flow open at a time. When composing to much the system starts freezing. In the end there is no more audio at all and I have to shut down the audio interface and Dorico and start it all up.
You might like to upload your Diagnostics file to @Ulf to analyze. Help > Create Diagnostic Report and send to u dot stoermer at steinberg dot de or attach it to this thread for him.
Cantai can take a while to render if you have a lot of lyrics in a long composition. You might like to give us details of your composition, how long is it, the number of singers etc.
Also, the Cantai team might like to have details. Reporting details at the bottom of this page.
What @arco said! Also a little more detail about your system — besides being an M1 Studio, how much memory, available storage space, etc. (If RAM has to swap out frequently with limited storage space, it will certainly bog down – if not fail altogether!)
I too experience performance issues. My MacBook Air which is known to not have a fan/cooling system is overheating when using Cantai – even after the rendering is complete.
I use it with NotePerformer which never had heating issues.
A long time Macbook user here with currently a MBP M1 and using Cantai on that system causes fans to run constantly. Since I do most of my main work on my Mac Studio M2Max, it’s not an issue for me as the Studio runs normally but it has a few more processing cores and 16G more RAM.
I suspect the reason for the overheat is that, if I recall correctly, Cantai uses primarily your GPU (graphics processor) for its work. And, if you just happen to be using the majority of your RAM, the GPU may have to be running more just to keep what you see and hear as near to real time as possible. I could be wrong…so please validate with someone else.
I can’t speak of MacOS since the plugin may work differently there, but on my Windows system there is basically no GPU usage as Cantai is rendering, so it seems to be using the CPU. It could well do something different on Mac because there are fewer hardware permutations to worry about.
I have a dedicated GPU, but I can tell you with certainty that Cantai isn’t using it. The GPU utilization is much too low when Cantai is rendering (around 1% or less), and the CPU utilization much too high.
I have run some local AI models that use GPU rendering and the usage pattern is completely different.
Furthermore, at least on Windows, the required DLL files for GPU rendering seem to be missing from the bundled ONNX runtime, which would force it to use the CPU.
We haven’t had time to get GPU and cloud use working yet for Dorico. It’s very much on our TODO list, however, once the dust from the Sibelius launch settles next month.
That’s what I figured. It makes sense to not include GPU rendering at first to put users on a consistent (albeit slightly slower) playing field with CPU rendering, which lets you focus your troubleshooting efforts on more global issues with the plugin (like the widely reported activation issues) instead of dealing early on with issues with specific GPU configurations or models.
Thanks for your comments. I will run a test on my machine using the standard audio interface of the MacStudio and besides the audio interface of my ZenTour. Keep you informed and make a log file. Grtn Jan
I can not get the render engine to start. The VST loads but doesn’t seem to start. I have loaded on my old WIN 10 machine and it works perfectly. I have hard time explaining so here is screen shot. Maybe some can tell what is not happening? Win11, 96GB RAM, Dorico preference set to allow connection.
I did import the playback template, went precisely through every step in the readme pdf, and waited many, many minutes. Still the vocal part does not play anything.
The disc space used in the Cantai popup window remains at 0 Bytes so I guess it does not render at all.
Track inspector shows correct connections, Cantai is the top VST (Noteperformer 2nd).
All other parts are played correctly via noteperformer although for some reason most level meters in the mixer show nothing.
I run WIN 11 Home on an Intel Core i5-8500 CPU at 2.80GHz, 8 GB RAM and 31 GB free disc space. The score is just a simple sketch of 90 bars and 14 staves, one of wich is an alto solo singer.
It works fine with a fresh score of solo singer + piano as long as it’s only a few bars. 10+ bars and it’t over.
I still have the same issue with all existing projects I’ve tried. Rendering only takes place when I create a new project from scratch. Curiously if I copy/paste the contents of an old file into a new project, it sometimes renders OK, sometimes partially, and sometimes not at all. What is perhaps most interesting is that with the original beta, nearly all projects rendered correctly.
I must say my interest in this is beginning to wane – perhaps I’ll come back in a few months when things should have moved on a bit. In the meantime, I’ll continue to use SynthV for my vocal projects.
we are now on 2.2.2 of the pug inn. I have to say, for the project I am working on. A mass with SATB. I have the four pieces in one project. The rendering takes very long and it didn’t do the bass and tenor. I have to rerender them. Also when playing arround it gives a click and now the bass and tenor are rerendered. It eventually works, but I notice I work with noteperformer and only when finished I set the playback to Cantai. What I also notice, is when you are working with Cantai. the plug-inn seems to claim memory which is not released. That eventually leads to clipping and break down.
I don’t think it has to do with my audio interface at this point.
So far, I am finding that it is easier for me to notate a piece entirely using the VST I have used “forever” before Cantai was even a thought. Then I go back and change human voices to Cantai. It still does some weird stuff though. Pronunciation of “alleluia” is “Al Lee Lou Ee Ah” because it apparently cannot pronounce or recognize “ia” as a diphthong when on a single note.
It took well over 30 minutes yesterday for a complete render of SATB of a 150 measure (bar) work that probably uses 90% of those measures (bars) for actual sung notes (130-135 mm). Then, when making phonetic changes to try to rid oneself of the odd sounding alleluias (8 measures in total for 4 voices) it re-renders the whole work again taking 30+ minutes to complete. I thought that the cache was supposed to minimize that sort of thing only re-rendering the change. Same with any dynamic change.
In reference to dynamics, the gain on these voices cannot be controlled overall which is a bit annoying. I am spending a lot of time trying to find the sweet spot using Dorico’s mixer. Adding some wet reverb and adjusting the sound space seems to help some. There really should be some master control with this VST, to adjust the overall gain for every voice at once. And, the dynamic range should be consistent with Dorico’s extended range of dynamics from pppppp to ffffff. Otherwise, there will be no consistency with matching dynamics of the voices with other instruments.
The end results have been OK but it’s a lot of work that is very time consuming in the bigger picture. All-in-all, though, Cantai is a decent product that just needs a fair amount of fine tuning, so to speak. Resource utilization, render time and pronunciation are probably the most important issues to address at the moment.
System specs: MacStudio M2Max, 32g RAM, 512g disk with 200g free less the ever expanding Cantai cache now over 1.8gb this morning
which unfortunately defeats the whole purpose of the exercise. If you simply want the best vocal rendering, there’s no doubt that SynthV is currently superior and much more flexible. Still, I haven’t ruled out buying a Cantai license (now that the price is again more sensible) as I really want this to succeed and presumably the more money they can rake in, the faster the much needed development can take place.
One of the things that worries me most at present is the cache. It clearly seems to be malfunctioning if it takes half an hour to render a few small changes. This I imagine must be sortable, though.
@dko22 I’m always apprehensive about purchasing a new sound library. Would you mind sending me a personal message that outlines your process using Synth V? As much as I think Cantai has the potential to do great things and far surpass other options, it just is not there yet and it’s probably going to take a while for that to happen - like after the initial bug reporting with the Sibelius version.
I’m a bit disappointed but remain hopeful. I just am tiring of having to think up workarounds and then add all the extra time for processing.
