Well, I have to disagree about the quality of the rendering. To my ears, it’s a long way from being good. Anyone who invests in it is going to have to spend a lot of money to generate a high quality product. It’s the last 20% of realism that is by far the hardest to achieve.
Cantamus sounds better for solo voices, IMO.
The key difference is that Cantai, unlike as such great Cantamus, will work directly inside the scoring program like NotePerformer. That’s the main reason I am so looking forward to Cantai very well knowing the understandable current beta stage limitations. The sound quality will improve little by little, I hope and am sure.
I agree with you. Very interesting technology and I’m glad to be able to try it out. But that amount of vibrato is impossible to listen more than once, and I can’t imagine working on my own composition with that kind of playback. The vibrato isn’t powerful, it’s… don’t know how to say… It’s hysterical. Hopefully, as time passes, it will be possible to get the voices if not 100% real, at least more restrained.
It would be interesting to listen to a 2-3 minute track in dynamics from PP to mP.
Yes, the approach is great, I agree! The background rendering you mention is already done by Ace Studio - but again, the quality of Ace Studio is not there yet.
Background rendering, yes, but ACE Studio doesn’t work inside the scoring program, either. 
Cantai will. Hopefully. 
Well, this is certainly subjective, and Cantamus is by every means a valuable piece of software, but currently there are limitations as for the number of simultaneous voices and the length of the piece, so it is another example of a promising app with lots of space for further improvement, albeit admittedly more optimised and refined already then Cantai. I think that the whole point of this thread is not to counterweigh comparatively the value of every present, future and prospective AI vocal generation app, which certainly will populate the landscape of music technology in five years time, but to consider positively and attentively the possible inclusion within this year in Dorico of a specific software temporarily called Cantai. If only because it’s just about time, that contextually aware integrated sound synthesis finally makes its long-awaited appearance, in the area of music notation.
I concur that the current vibrato is excessive, and solely catered for high-octane passionate operatic music. The developer is actively working on training a different vocal model with a more linear delivery, which I have requested myself among others, with a melodic approach suited to both contemporary music and Baroque/Renaissance polyphony. I think that the current Cantai’s virtual singers aim at emphasising human expressivity, to differentiate the software from other rather flat-sounding vocaloid-like algorithms. I would not use myself the current operatic voices of Cantai for my compositions, because they would sound completely out of place for my taste and style, but I am supporting the development because it promises to deliver soon different vocal models, and a completely user-customisable singing engine.
Hey Dorico, I am in the Cantai adventure too
and I don’t want to be obliged to go to go to Musescore or Sibelius
But I’ll do it if I need to
Welcome to the forum Alexandre.
Jesper
I’m wondering, will cantai support microtonality with dorico (like noteperformer)?
well exactly. Cantamus is far better for chamber scale choral music. EWQL Hollywood Choirs (and for boys voices Symphonic Choirs) sounds much more musical despite it’s limitations than anything I’ve heard from Cantai so far. It’s not the technology which is the most important – I’m as excited as anyone over what Cantai are trying to do – it’s the quality of the voice models to date, especially for choirs. I can see soloists , where there is little competition and one two more effective models, being the real initial breakthrough.
I hope this is not too much “advertising”, but anyone interested in Cantai might want to see and listen to this (YouTube). That’s what – and later more and better – you will get directly inside and from the [Cantai-supporting] scoring program of your choice.
New demos of cantai. These were all rendered in the web interface then mixed in Logic Pro.
Our most pressing bug right now is the remaining pronunciation issues. Hoping to wrap most of those up today and tomorrow.
Very impressive!!
It all seems to work, there’s a (beta) Render button on the website now.
I’m an early adopter hopeful, yes impressive, at least for me.
For some reason my password from my purchase did not work so had to reset it.
There are different ways to get it working from Dorico, including this one described in another thread.
For me, I exported musicXML from Dorico (edit: current version 5 of Dorico does not export midi with lyrics which Cantai web render needs), imported into Musescore. Tidied up there (I had to separate my divisi as it does not do this), export midi, on the Render page render it. Two tracks appear, one dry, the other based on reverb settings you can set before rendering.
Richard demonstrating the process (Musescore to Web render) and with Sibelius (linking to them here although the links would be elsewhere I would think.)
I imported the .wav files into a DAW and added my own reverb. (No instruments, just singing.)
The Please Read box and Guidelines on the render page gives very useful and important information.
Slowly various more features and voices will be added we are told.
Richard’s (today’s) YouTube update here titled “We’re LIVE and this is the WORST it’s ever going to sound!” specifying its only at about 25% quality.
One thing I notice now, is at times, the singers are not taking any breath where normally you would expect them to (despite the continuations of notes, ties, slurs into one another). I will have to check slurs or if I need to shorten note lengths just a little (easy in Dorico with the key command) or add breath marks (I presume Cantai sees them or will do, I have not yet checked.)
I wonder if breath in (sounds) will be (although not sure how) added or perhaps they are already but I have not listened carefully for any.
I also noticed some of my pp dynamics were too soft for what I wanted, so deleted them for the render which worked for those sections. I’m wondering if a (midi!?) compression or expander might be useful, where the Cantai engine has a different (dynamic range) conception of the differences between dynamic markings. Presumably or obviously a softer pp has a different timbre than at an mf for instance. Something more to test, check out as I progress.
Those of you who are using Cantai:
Can you explain the mixing process, if there is one? Does the user have the option of alternate mixes on a work? For instance, the whole reduced in volume overall, except for the choral tenor part - for note-learning for a chorister? Is that an option?
Does this answer your question? Separate .wav files for each singer, mix in DAW.
(But I think you are envisaging importing one composite audio file of the singers or multiple audio files, one for each singer back into Dorico, then mix those?)
Perhaps with the VST when it comes, there might be a mixer for each voice, or perhaps not, as it’s controlled by the assigned dynamic markings only.
Perhaps Richard will enlighten us in due course.
Thanks much for this link. With files pulled into LogicPro, part-predominant rehearsal trax would be a cinch to mix.
Yeah exactly.