Anyone here has heard/read or knows more about this project/software? Opinions? Expectations?
Ohhhhh don’t tempt me.
I’m skeptical until I see it myself. No offense intended to the creator, but many of us have been seduced by VST demos, only to find that the tweaking behind them was Herculean.
That said, this would be a massive step forward.
Interesting, I must say.
If it works like it is advertised in the video, that all the information is in the score, then it would be like NP for vocals.
Exactly… I’m extremely tempted and extremely skeptical at the same time
That’s my hope… NP for vocals.
Yeah, I’m gonna buy this. Just for fun and for 125$…
Looks like they are one step ahead of my beloved SynthV, as they can take the lyrics from the score itself, which is huge.
Even though I usually do not write for “classical” vocals at all… But I’m a sucker for cool tech!
I’m intrigued – but Dorico doesn’t pass lyrics to the VSTs, so how does it work “with no interface”…?
Very good point… On discord it still feels like early days, to be honest, but he talked about a free trial version and a generous refund policy… The whole thing seems to be centered around MusicXML at the moment.
Actually, still haven’t gotten the app yet, and am not sure I will after all…
I think, the statement “just like NP” is fairly misleading, since there seems to be a good amount of offline processing involved.
If they could get the lyrics transferred, then gave it a second or so to render (SynthV takes less than that!), then we would be talking…
In my opinion, this tech, classical or pop needs to be:
- Locally computed,
- Hardware accelerated (CUDA/tensor/…),
- Close to realtime, give or take a few seconds,
- And taking the lyrics from the score.
SynthV has 1. and 3. covered, but lacks the classical singers and choirs, likely due to its roots in Japanese musical pop-culture, just like Vocaloid.
Let’s wait and see what gives, the future is here already anyway!
B.
Where can I find the Discord channel/server?
If it really, really works like the inventor says it does, then…wow…
Still hard to believe. But so was Midjourney a very few years ago…
It looks like the developer answered this question in a comment to the YouTube video:
So they have to build a “low-level integration” with Dorico, and until then, it will require a step of exporting a MusicXML file that Cantai can read to get the lyrics.
I’ll do my best to answer the questions down below (I’m the guy developing cantai), and any new ones here.
My approach from the beginning was to emulate the transparency and score-based workflow of NP. So that’s just what it does. Reads the entire score: notes, lyrics, dynamics, slurs, performance notes, etsc and builds the performance from those.
I am integrating it into Dorico using their API, so that it has much more low-level, complete information than a VST would have without it.
Odd that Chinese and Japanese were thought to be more urgent to integrate than French, Italian, German, or Russian, which surely make up a larger part of the classical vocal repertoire?
I’m watching this thread, and Cantai itself, very carefully. If there were an app that could “sing” and interpret choral music with a genuine choral sound (meaning NOT pop or even operatic), the market for it would be huge. There are many, many university and church musicians, as well as choral composers and publishers generally, who have long dreamed of this.
I bought it yesterday…
Let’s see.
Richard (the developer…) hinted on Discord about a local installation without the need for cloud processing in about 6 months time.
I’m not really worried about off-site AI, and it’s likely a more expedient way to do the heavy lifting at the moment, but the install size is around 1 GB, so there is at least some AI data present locally.
Exciting times!
Benji