In a work for piano and voice I assign (route)
and it all works great. ** 
**
But I don’t know if VST’s are multithreaded. So in a work for SATB and piano what do I do?
Do I create:
To the right person the answer is obvious but not to me! 
Each voice needs its own Cantai instance.
If you use a playback template that has Cantai and then NP as a fallback, this should all happen automatically for you.
You can download the Cantai playback template from here.
And this video shows you how to create a composite template.
Ah thank you for your reply.
If I load this template will it effect All Dorico projects or just one? I do hope its just one.
Currently I have a massive 45 min project with many flows. When I want to check vocals in detail I
(Other wise I’d be waiting for Cantai for ever).
A playback template only affects the current project.
(EDIT: I would add a caveat. The file’s current sound setup will be permanently removed and replaced by the new playback template. If, for some reason, you want to retain the current sound settings, save a copy of the file first!)
GREAT - the right answer! 
This works great, it takes checking of vocal parts to a completely new level.
However, it appears that the template including Cantai is applied to the entire project (not individual flows within the project). In a long project I guess Cantai will then attempt to render the entire project (in my case ~40 mins of 4-9 voices) not just the flow we work on. We’d have huge waits.
I have 18 flows (‘arias & choruses’). They are relatively independent.
Today - Is this the best work plan? If there’s a better solution I’d love to know about it.
‘Just for information’ to anyone who is composing a vocal work with Cantai.
Dorico has superb facilities for exporting and importing flows. Because of that its easy to work with each flow independently in its own project. For me longest flow is around 5 minutes, but most are around 3 minutes. I have 5 soloists, SATB and piano but most flows don’t use everyone.
I compose with everything set to playback via Note performer.
I have a hot key combination to switch between NotePerformer for everything and Cantai for voices and NP for keyboard.
After I switch to Cantai I exit and restart Dorico (which only takes a couple of seconds). Then you have to wait for Cantai to build its caches. Typically this is around 1-2 minutes - I have ‘Task Manager’ visible so I can watch the Cantai application do its stuff. When its finished I can audition the whole piece and make (text) notes on paper. (Dorico also lets us export MP3 at this point. We can use that while editing the score in NP output mode)
I haven’t had much success changing anything in the score while in Cantai playback. Obviously making changes is invalidating parts of the Cantai cache but it’s difficult to predict the exact effect of that. Sometimes some of the voices drop out of playback.
So for now I would (personally) recommend doing the actual composition in NP and using Cantai only as a wonderful audition tool.
I’m sure the Cantai system will get much better with time; but it’s already very good.
Even so this existing workflow is superb; simply being able to correctly audition the vocal parts is a game changer; it should save some unpleasant surprises when scores are given real performers.
Glad you are getting into Cantai – I am impressed and frustrated in about equal measure because there are so many bugs and issues at present – the future looks bright, though. On some specific points:
Many have reported half an hour or more. It depends entirely on how large your project is. The choirs seems to take much longer than solo voices (though they are currently far from ideal so I wouldn’t use them anyway). That’s for the initial rendering, though to be clear – after a project is rendered it should only take a few seconds to re initialise when opening in my experience.
I wouldn’t dream of using NotePerformer for auditioning when there are far better tools in terms of sounding something like actual instruments-- above all for vocals. But there are quite a few NotePerformer devotees on this forum and I find it still very useful in combination with third party libraries through NPPE - alas no longer supported.
I actually am fairly happy with the update speed when making changes-- normally no more than a few seconds.
not when most existing projects don’t render at all as is the case for me and quite a few others. The only workaround is exporting the vocal flows into a new project and then recombining in some way with the instrumentation (easiest in Cubase for me)
I have have never had that problem. Nor have any renderings taken longer than half the duration of the music for an average of about 5 parts.
I have no special insight into the internal workings of Cantai but there is a cache directory here:
C:\Users[Your User Name]\AppData\Local\Cantai\local-server\Cache
that fills up with files. I’ve occasionally cleared that directory myself when Cantai is not running.
In time I’m sure Richard the Cantai developer will create a PDF describing how to use it most effectively.
Amongst many other things I used to write cache systems commercially as a software engineer -‘housekeeping’ - The clearing up of old cache files is something which usually gets a low priority and hence often does not work correctly. . . (That’s why Windows computers are full of Microsoft’s cache files…)
Existing projects not rendering (or only partly rendering) is a known problem which the developers are actively looking at – including one of my files I sent them. In due course I assume we’ll know why some people are affected and others not.
Clearing the cache means clearing out all the audio information which has been built up to enable fast rendering when re-opening a project. I thought myself that possibly something in the cache might have been causing the problems so tried in the past clearing it once or twice but it doesn’t help.
As for the time spent in initial rendering, it’s obviously system hardware specific. The choirs are notoriously slow (and can be speeded up a little by reducing the number of voices). For the most part, I find the performance acceptable – some others beg to differ!
I’m sure that when the Sibelius version is out and its initial teething troubles are sorted, the developers will be able to tun their full attention to sorting out some of the more serious bugs in the Dorico release.
Actually you are correct. This works better than I realised. Especially if I keep an eye on the Cantai process in task manager and wait for it to subside before auditioning anything.
But then just occasionally rather than ‘working better than I realised’ it just doesn’t work…
I’ve also tried little bits of vocal composition in a small ‘scratch pad’ project. That works well provided you don’t switch too quickly between two Dorico projects. As you might have experienced rapid switching will cause Dorico to hang.
I’m doing all this on a medium power gaming laptop. I did wonder about upgrading to a top level laptop - where presumably everything would happen much faster. But it’s a lot of pennies and so much wasted time setting up yet another machine.
Eventually if Cantai starts to use the GPU we should enjoy a vast increase of speed.
I agree on the GPU – although many computers, especially laptops have only a weak GPU so it’s important that software development should no rely exclusively on it.
I also agree on switching between Dorico projects – not that smooth at the best of times and with Cantai involved, the app freezing is rather common.
On my Mac (M2 Max chip), it appears that Cantai is using the GPU somewhat but nothing near the potential use. I’m not entirely certain but I sense that part of Cantai may still be accessing the internet if available for some of its work. My ISP router went down briefly and Cantai slowed down at the same time. It could be coincidental but…
Initial renderings can take quite some time in multi-flow projects but I have experienced a significant decrease in time with notation changes in voices the more frequently Cantai must re-render. I try to only make changes in one voice at a time because everything slows, as I would expect, with multiple re-renderings occurring simultaneously.
So to the original question, if not already answered, you need one instance for every voice that Cantai “sings.”
I see a similar result in windows - just part of one of GPU’s is used slightly.
I wonder if right now, even though it is touted as locally rendered, that the AI is reaching out to the web for help…
I don’t think so. I think it is checking the licence. turn the Internet off and you will see it does not run.
My system is quite workable for 9 voices in a 4 minute work. Rendering time is similar to duration; cached changes are a bit slow
Its all good provided I only use solo voices as loaded by the Cantai template. I tried the choral voices but could not wait for the result.
My system has a Ryzen 5600H processor and 8Gb memory. I could buy a system 3-4 times faster (The web site “passmark” is our guide to this, you can check CPU chip performance and then discover which systems will have much faster performance).
Before I do that - anyone know how much this will help?
I’m using a MacStudio M2Max and it is slow. So, an advanced, fast processor doesn’t make much difference. It is supposed to be using the GPU more than CPU cores by my understanding.
I’ll guess GPU is yet to come. If it happens it may be a 'Game Changer’ because the GPU is designed to do simple things in parallel, which of course could be great if it’s doing a choral part, or even just SATB all in one overall pass.
Its hard to compare machine performance: especially as the MacStudio M2Max family comes with a variety of different CPU chips. ‘Passmark’ lists these:
So, it’s really the number of cores that makes the difference (assumption based on the data you shared). Mine is the 12 core 3680 MHz and I have 32 GB memory and, if I understand Passmark correctly, your CPU score is 16409. Since mine is significantly higher score, it would appear that a faster CPU does not, indeed, make a difference with the time required for a Cantai render. I do not believe that Cantai is really taking full advantage of modern processors so I would not be inclined to upgrade based on current Cantai performance.
