The significance of Cantai

Dorico
1

I know Cantai, in its current form, is not for everyone. Clearly it is the beginning of something, not a fully polished product. But I think we should take a moment to recognize and thank Steinberg for being willing to to invest significant development resources in something that may not have a direct pay-back to Steinberg but is clearly an important step in the industry.

These are not easy times for makers of music products, particularly on the software or creative side. I think it is noteworthy that Steinberg is willing to embrace AI-related things rather than to try to become insular.

Open systems are scary. Many companies naturally gravitate toward proprietary things. But it is clear to me that the real power of technology in the music field comes from openness and the ability for products to work together. I bet there is a lot of soul searching within Steinberg and its peer companies on this issue. I, for one, appreciate their choice to come down on the side of embracing connectivity.

2

I know a lot of people are poo-pooing cantai in its current incarnation, but the fact that any automatic singing is happening at all is really quite something. More progress has been made by cantai in a short while than by other VST companies that have attempted to do something similar (albeit less automatic) for years. It certainly feels like we are at the bottom of the hill looking up, not the other way around, and I find that exciting.

By consequence, I’m grateful to the developers of Cantai, as well as the Dorico dev team for making sure that they can play nicely together. Things will only improve from here.

Little aside

As an aside, and not meaning to derail the thread, I also think the alarmism seen in a few places on other social media platforms about “AI TAKING JOBS AND OUTSOURCING ART!!!” are totally missing the mark. This is a compositional tool, not a replacement for real people. (To wit: products which aim to be replacements for real people DO already exist!) But to have a new tool to help composers audiate and iterate their compositions is a good thing. Noteperformer, lovely though it is, will never replace real string players making real recordings. Cantai will never replace real singers giving real performances of real works. But it sure is nice to have something to show for it in-between such real performances.

3

Not at all. I think it is all related. As an aside of mine, last night I had a rehearsal with a jazz big band. For our next show, I have been thinking about doing a piece that I “created” on Suno. It is not a great creative masterpiece, but it is a fun little thing.

As this show will be in the largest jazz club in our region, I know there will be some in the audience highly hostile to AI. I might get boo’ed off the stage, but what I am attempting to show is that AI can be a useful tool. The tune is catchy with some goofy lyrics, but that’s what I expected when I asked for a song about “unsuccessful fishing in the Caribbean islands.”

And the point I will try to make is that a live band “owning” this song is way beyond anything we will ever see from a computer program. I did a sketch arrangement for the rehearsal, and the band loved the tune. It is sort of a “Tower of Power goes to New Orleans and hooks up with a Jamaican ska singer high as a kite” vibe.

And I did use Cantai as I was doing this sketch. It was nothing like a real singer, but still pretty amazing. I will experiment with some backup vocals using Cantai. Too bad they can’t do polyphony yet.

4

@turingopera outlines the process here

5

I concur, @cparmerlee.

6

I wish Cantai well. Unfortunately, I was about to purchase it at the pre-order price but missed it by literally one day. If it ever goes on sale, I will reconsider. I was hesitant until I saw the Steinberg video.

7

If I understand it correctly, from their FAQ page:

it says this:

Does one license cover all integrations?
Yes. A single lifetime license covers Dorico, Sibelius, and MuseScore Studio.

I wonder if you purchased the Sibelius version, the license key would work for Dorico anyway.
You could ask Cantai, unless someone replies here it does not.

8

Their pricing page states Sibelius version pre-order $150 and Dorico $299. I’ll wait and see.

9

I was wondering about this as well tbh. I feel Cantai is too crude at the moment to be of any use but it’s not completely inconceivable that it might have made sufficient strides by the time of the release of the Sibelius version to invest in that license. After all, in theory a license is for all platforms.

On @cparmerlee’s initial point, I’m not sure how significant Cantai is in itself. it’s actually relatively late to the table in vocal synthesis which would explain why it’s so far behind the likes of SynthesizerV in terms of expressiveness and sophistication. After all, Dreamtonics were working for 4 years on their set of choirs which is the current benchmark. It’s hard for me to even consider Cantai at present when I know what’s actually possible. Having said that, the idea of integrating directly into notation software, rather than through an external interface is new and potentially a big time saver when it starts to work properly and considerably faster. And naturally if the vocal models improve and have their own clear personality (there are some signs of that already), then I’d be only to happy to invest in what is still a market with relatively few competitors. However, if the ambition for Cantai is to go no further than a basic, NotePerfomer style proofing tool, then you can count me out. I’d like to think they want to achieve more, though.

10

I don’t know how sophisticated his licensing scheme is but it might be that the pre-order version for Sibelius will not activate until the Sibelius release date.

11

I can’t speak to their ambition. I assume every supplier wants to take it as far as they can. I think the NotePerformer comparison is exactly the case. There are better sample libraries than NP – Indeed, NP tried to work with them and several of those suppliers foolishly (IMHO) rejected that. It is their loss, because they will be gone in a few years because of AI.

Cantai and NP both offer drop-in-and-go ease of use. You can get better results if you want to spend > $2000 for sample libraries and hundreds of hours trying to integrate and tweak that yourself.

It really is two different use cases. The NP / Cantai users are looking for a way to proofread their material with a degree of realism, and to produce renderings that won’t fool anybody but are very useful for rehearsal purpose, or perhaps to pitch a song concept. On proofreading, I often find myself adding slurs and articulations in order to get NP to express what I want, and almost all of those markings are good for human musicians too. And now with Cantai, I have already found misspelled lyrics in my parts because the “singers” sang what was there, even when what was written was rotten.

BY the same token, NP’s experience dealing with the various sample libraries suggests that there may always be a quality gap. But the big difference is that NP and Cantai will still have a place for many years due to their ease of use and tight integration. The expensive libraries are not long for this world, as they will be overtaken by AI before you can say “Justine Wilson, Grimes, Shivon Zilis, and Ashley St. Clair” (the 4 women who have carried Elon Musk’s babies – Happy Mothers’ Day.)

12

Perhaps you know something the rest of us don’t. Certainly NP was put under pressure from at least one developer but it may well have been just one. Both Spitfire and VSL came out publicly as saying it was nothing to do with them. To me, the raison d’etre with NP was the Performance Engine extensions though of course not everyone agrees. And with native NP , I agree that proofreading and mock ups are two different things. In general I’m interested in the latter.

And perhaps indeed the sample libraries will indeed be replaced by AI. But then perhaps so will NP. And Cantai with the amount of time it’s taking to render the text – when it does so at all which for many it doesn’t-- makes it very far from comparable with NP as things stand.

13

I have no special knowledge. My recollection is that Arne at least implied that he was getting grief from more than one library company. I would have thought that if it was only one company complaining, it would make sense to drop support for that one product and continue with the others.

Be that as it may, I think life will be very difficult for the companies asking $2000 for some samples when we are seeing AI progress so quickly. IMHO, it would have been far wiser for all of them to have collaborated with NotePerformer. There is no way I am going to buy one of those products with as much tweaking is required to get good results. But if I could get good results with their library quickly using the NotePerformer integration, I might spring for one of the expensive packages, even as a casual user. My work is almost all in pop music, so none of the packages were really an option for that.

I bet before too long, it will be possible to send a MusicXML file, along with some commentary about style, to some AI program, and have it render very nicely.

14

well this was my favoured path (and one I still use with a couple of libraries). But it is indeed not unlikely that these libraries will all be replaced by AI at some point before too long. I can’t imagine myself buying a new large symphonic library these days, though personally I quite enjoy the work creating Expression maps for them.

15

me too. I often go back and revisit older libraries that I created expression maps for and refresh them with new and improved knowledge. It is an amazing feeling bringing new life to long forgotten libraries.

16

The price is now $149 (May 22, 2026), with lifetime updates

17

Yes thank you. Got it. I think Richard listened to his users.

18

FWIW, I’m finding Cantai 2.3.3 to be very good. I was stuck on 2.1 because of the apostrophe problem. With that problem fixed, I can hear they have made a lot of progress between 2.1 and 2.3.3, in less than a month. That’s really quite impressive. It still has some artificial-sounding phrasing, but really, as a music demo, it far exceeds my expectations. I am not running any complicated scores with repeats, multiple verses, meter changes and all that, so I’d expect there are still some bugs to be uncovered and worked, but I can’t find anything I don’t like about this.

I needed to make some adjustments to an arrangement of a Michael Jackson song, so I activated the template and put “Emilio” to work. Honestly, that doesn’t sound so much different from Michael jackson himself. I mean, it wouldn’t fool anybody, but it really is a very useful playback for my purposes.

=====

A new day and I’m working on a different chart that has first & second endings and DS al coda. Some room for improvement here.

To elaborate on this, the only big bug I stumbled on was some messed up timing coming out of a second ending. No matter what I did, the next measure , the singer was 2-3 beats late, but then it adjusted by the 3rd measure after the second ending.

But there are many cases where Cantai doesn’t seem to be doing anything. Maybe the VST UI will say “rendering”. But the status line at the bottom of the Dorico window often shows nothing, even if the VST says it is rendering. So one is left to conclude that Cantai is dead.

However, sometimes it eventually comes back to life on its own. Other times, I can revive it by clearing the cache without restarting Dorico or rebooting. These are rough edges that will make new users think the product is not ready for real use. In fact, with patience, it is producing some very useful results.