CeVIO Kafu announced for CeVIO AI (Early 2021)

[VocAddict]

The problem with those demos from 2018 is that the engine wasn't optimized for being able to sell to normal consumers. It took 5 hours to render a 5 min song. So they had to compromise/balance it to get it to a point the software was better than CS7 but actually usable for most people.

I wish we knew for sure if we could even TUNE in CeVIO AI. Because they had to edit the demos they showed with Melodyne, so that sorta sounds like... no parameters to me??

They really are gonna make us wait until the last second to see the actual software, aren't they? *sigh*

I wish we knew more.

Totally forgot how long it took to render that rip. That does sound like no parameters to me as well, kinda like Neutrino I guess? At the bare minimum, they could allow pitch editing. If Microsoft could have done it for their AI cloud singing synth, I'm sure those behind CeVIO could do the same.

 

[lIlI]

A few positive notes: I'll say that elements of her tone sound amazing. The engine noise is much smoother and more pleasant, she's also significantly more expressive than previous CeVIOs. Just needs a different voicetype for me.

 

[Stardust]

Like everyone else, I am disappointed and confused.
Her VP has a very unique voice, but the demo showed a voice almost completely different from her VP. If it is just Gender parameter, then that is fine, but it seems like a bad idea to use a altered version of the voice for a demo when they should give us a tuned or even raw version of the voice. I do not mind the voice if that is the final product, as it is a fine voice, but it seems like a disservice to the VP and what her fans were expecting.

I also agree with |I|I. I can hear an improvement on how CeVIO is handling voices now, especially CeVIO's infamous engine noise issue.

 

[Twillby]

I’m weirdly optimistic about the situation? Considering the first time I listened to Kaf after the news broke I went, “Nuh uh this will not be a voice for me.” I do admittedly like the edited samples more so I’m hoping it’s a case of questionable tuning but I feel like it might be workable either way (even if it means having to theoretically pitch a render down every time X’P).

Maybe it’s just me being happy to have SOME kind of news For a while there I was really worried we’d get a surprise release without the slightest scrap of demos (though it makes me wonder if she’s releasing alongside Yukari/Flower... considering I thought they were going to be the first banks yet have no demos so far)

 

[msmsmsm]

I had a hunch her voice would sound a little strange when synthesized because of her odd vocal manipulation and pronunciation, but honestly, I'm a fan! I've said it before, but I'm seriously rooting for CeVIO in their AI endeavours. I agree they could've demonstrated her vocals more clearly in the demo though.

This also gives me hope for Flower and Yukari, knowing their pronunciation is more standard and will most likely respond much better to the engine than Kafu.

 

[morrysillusion]

after knowing what Kafu sounded like and then hearing that first singing sample, i was like...no dont like that. her real voice is interesting and i like the potential it has to sound in a lower range- that synthesized sample was....not that at all. i did like how it sounded in the edited pitched down version someone did and i hope that maybe they just used her in a different range or something. i dont think that sample sounded much like the VP at all and i really wont have any interest if she fall into that higher, more obviously cutesy tone tbh.

i will say its nice to hear a more new sample of the engine, i dont think it sounds too bad. she does sound a little muddied/muffled on some sounds though and idk if that will change or improve... but the lack of info still, mostly regarding how it works (how does tuning work? can we tune at all? considering the way many AI based synths are working, i am afraid of that limitation) is keeping me from wanting to put money into it and i would like to know...to some extent, so i dont end up disappointed that it does/doesnt work a certain way on release!

 

[YOYo_MAMA]

I'm not surprised they tuned Kafu to be that high. It's almost like a required trait that female vocaloids(?), vocal synths(?), at least the Japanese ones in my observation, to be high pitch e.g. Miku, Rin, Una, heck my fav, IA, sounds infinitely better if she was tuned low like a soprano Opera singer but some of her most popular songs are her tuned so high it could brake glass. Vflower is meant to be androgynous, neither female or male but a bit of both, so she can be deep pitched without straying away from her intended design.

To be honest, I'm neutral on the Cevio bank of Kafu, even though it sounds like Aoki Lapis' but more nasally pitch. I don't hate it but I'm not super excited for it. If i see it I'll be like "That's Cool", then move on. Or maybe since I'm not into Vtubers, and would rather wait till the actual release and usage than just 1 video trailer.

I'm more interested in the technological development on Cevio like how they finally removed that engine noise that makes the songs sound like it's under heavy autotune. And how they intended to capture the whispy trait that's also in her original VB, that's the unique trait that should make the Cevio VB stand out from the large roster of "moe" VB's that already flooding the market.

 

[Exemplar]

 

[mahalisyarifuddin]

I love type A. Voted!

 

[Blue Of Mind]

I was just briefly ranting about this on the VV Discord, but Types B and C sound awful. But apparently, Twitter is favouring Type B?

 

[mahalisyarifuddin]

I love that amanokei just broke the code

[PS] I just tried myself but if you want to have the type B voice, just do the other way around. It worked!

[PS] Well I love this girl! But after things @YOYo_MAMA just said, I kinda afraid that this really potential voice will ended up "tragically" the same like IA.

 

[uncreepy]

I didn't realized VocaSphere already translated part of the survey, but I did, too.

Anyway, I read the comments on the poll anouncement and talked to a Japanese user/got talked about by a Japanese user and this is the summary of what is happening:
Kaf has many fans, many more than are producers. Her company is worried that fans of her voice don't want people to be able to make songs with her voice, which is why they made TYPE B. Japanese comments on Twitter agreed they thought TYPE A sounded like Kaf, no one really liked TYPE C, but they thought TYPE B was a compromise between A and C. Some users explained that they thought only Kaf should have Kaf's voice, so they were worried.

Non-Japanese synth fans overwhelmingly didn't know who Kaf was before this announcement. But some did know about her and like her voice/want to have it. Some Japanese users think that western fans only want hyper-realistic voices. But I explained on Twitter that it was more that "Why make an AI voice if you don't want it to sound like the voice provider? Why did they pitch up Kaf's amazing voice?"

So I guess I understand now why they changed her voice to be higher in order to "protect" Kaf while also pleasing her huge fan-base. So if it ends up that TYPE B wins, I guess we know why/probably can't complain too much because while it is disappointing, the argument does make sense for the most part. (Except for "Why sign up for an AI voice, why not just make a character voice bank?")

 

[mahalisyarifuddin]

Well it's not actually the case of we want ultrarealistic voice -it's why Kanaria's "King" or similar type of songs got so popular even outside Japan, it's just, we just already have millions and millions of high-pitched voices and it's so boring if we have another clone. Not to mention that the high-formanted type lost the characteristics due to the formant alterations. And if the CeVIO AI have formant alteration feature, what's the point of making high-formanted type the default?

But yeah, it's predictable if Japan peeps still want high-pitched voices.

 

[VocAddict]

I find it concerning they consider Type B a compromise like... doesn't the vocal synth scene have enough of those kind of voices? And I agree, it makes no sense to be making an AI synth and not have it sound like its voice provider, you would think that's the point of the whole thing but apparently not.

 

[Kona]

I liked Type A much more, and am hoping for it, but I understand the sentiment behind Type B.

Personally, this kinda reminds me of Fukase where I used to hear about a lot of the cinpany trying to protect real Fukase and not have people say Vocaloid Fukase is the real one—which I might be totally off on but that’s what I remember somehow. Kafu is in that gray area too, except, in my opinion, CeVIO is far more capable of realism in the right hands than Vocaloid. And if Kaf is as popular as I’m now seeing, I wouldn’t be surprised if someone tried claiming real Kaf was singing their song/cover. Could make Kafu’s license just as messy as Fukase’s...

Either way, I loved Type A and I can learn to love Type B. Actually hearing samples makes me pretty excited for CeVIO AI, even more so than before.

 

[lIlI]

Really interesting, I can see what the company is trying to do. It sounds like:

Type A: Natural
Type B: Moe
Type C: Moe + robotic

These are all types of voices that have their own popular niche within the vocal synth community. Not too surprised that Type B is winning for a Japanese voice, as moe voicetypes have been the most popular on the market since Miku.

Producers and Kaf fans are also going to want different things. Kaf fans enjoy the official media and personality that's been created for her, and they're used to it. They don't want new 'fanmade' content interfering with the style and story they're invested in, or replacing the current Kaf in the public's perception.

Producers want an authentic representation of the voice that initially peaked their interest, and something that stands out from their current options. Hopefully the 'shift two semitones' trick can be used by any buyers that want her natural tone.

 

[Stardust]

While I prefer Type A over Type B, it is interesting hearing a moe-type VB on CeVIO AI. It seems to handle them better then other synths to me, as it does not sound as forced, for lack of better words. That may be related to the recording process however, I am not sure.

I also feel Type B is pointless since many will most likely do the semitones trick to achieve a Type A effect, so I think the company should just release Type A. However, the points mentioned above are most likely why we will end up with Type B. I do not know why they even went with a poll considering it seems Type B is the safer option, and the one they would want to go with.

As for Type C, I do not know what to make of it. It sounds like Type B with techno tuning and mixing, something anyone would be able to achieve. I am not sure why a tuning choice would be on the poll.

 

[Rylitah]

I understand the sentiment - in the end, if Kaf's original fans overwhelmingly want Type B more, then who are we, mostly people who had never even heard of Kaf until this was announced, to talk over them?

I probably personally won't be listening to much of Kafu if Type B is going to be her default voice though orz, but at least it'll be that most people are happy with that compromise if it comes to it.

Also, I get the "we have too many high voices" sentiment too, but Japanese vocal synth fans don't really care. Kafu is pretty much mostly geared towards Kaf's original fans, who... most probably aren't vocal synth fans anyway? So they might not even know more than Miku, or know of how many high pitched vocals there are in general. So while it's a valid complaint to us, people who tend to follow every vocal synth release ever, this is mostly a different audience altogether.

 

[___]

In the end isnt type B just practically Kafu on the gender parameter? gbhnjmk I'd find it amusing if they went out of their way to do all this stuff only for us to be able to achieve type A only through fiddling with the gender parameter ( that is if Cevio AI has it but cmon...thats like the bare minimum expected from any voice synth engine )

ofc this is only hypothetically speaking

 

[Prism]

Both a and b seems popular. I honestly would of wanted a D the had her forment lower. I hope A wins so we can have the most natural bank. I could see all of them being appends tbh. I feel like c kinda ruins the purpose of ai.