I have a lot of hope for Yukari, she should be okay... It's strange because the comments seem to be eating it up, but in the overseas fandom I've not met a single person who's happy with this voice, even among the people who I know that are fans of Kaf...
CeVIO Kafu announced for CeVIO AI (Early 2021)
- Thread starter uncreepy
- Start date
-
- Tags
- cevio ai kafu virtual singer vtuber
There’s cutesy and sweet, and then there’s... whatever THAT was. And this is coming from someone who can’t get enough of really cute and sweet vocals. But if it’s getting good feedback...
It sounds like she has something stuck in her throat. I feel like anyone who was planning on pitching her voice to get KAFU to sound like Kaf will probably not be able to do that anymore. Yikes.
Correct me if I'm wrong, but aside from otaku types who eat up anything cutesy, doesn't Japanese culture usually discourage direct/harsh criticism (i.e. if they criticise something, they usually do it in a polite way)?
Hearing that demo is so disappointing. I've pretty much lost all interest in Kafu now despite the cool design (unless someone figures out how to use her in a way that isn't obnoxious).
Trying to read the comments on the YouTube video made me wanna barf... everyone was like "Amazing!" They only had praise for it saying it was cute and wonderful and I felt like I was legitimately in the Twilight Zone. Same with Twitter comments on the announcement/retweets of the announcement. They thought her breath sounds were so amazing you couldn't believe it was vocal synthesis. It seems like comments that were in foreign languages (ex: Spanish) hate it just like Western fans and only Japanese fans like these results. I guess we'll know for sure if they're only talking praise to be nice based on how many people buy/actually use KAFU.
Final voice aside, recording a synth of a popular figure and then messing with it to the point that it's not even recognizable as the person it's supposed to be defeats the entire purpose. She's purely being marketed off her character and not actual product quality, and that really leaves a bad taste in my mouth. I'd honestly be furious if I was a Kaf fan prior to this.
I know I said before that I understood the final decision, and I still do, but... this wasn't the way to do this.
I know I said before that I understood the final decision, and I still do, but... this wasn't the way to do this.
This is just awful. They should have scrapped this bank and started over to get the effect they wanted. :(
A lot of the Japanese comments said that hearing the demo, you could immediately tell it was Kaf. They are calling her voice little bird or little chick (I can't remember for sure and I definitely don't want to double check or I'll die inside again).
The thing is, hardcore Kaf fans DID want this voice type because they were protective of her voice. So they are really happy right now (the Japanese fans, anyway).
I agree - if they wanted a cuter version of the real Kaf, the VB should have been recorded that way from the start. Editing the original voice to make it sound cuter just made the final product sound way worse than expected.
This is going to be super speculative and coming from a limited point of view so if I say any bs pls correct me.
But I can hear obvious drop in quality, it is important to note that my japanese language skills are limited but she does sound very muffled to me, beforehand I could make out the phonemes she's singing like:
I can make out that she sang "zenbu"
But here it straight up sounds like gibberish:
And one thing I am speculating on is that maybe since they're not pitching type A after rendering it but actually..the vb itself?? It's messing with the AI? Because here the tuning sounded a lot more messier, it was disgustingly pitchy and I could hear errors. It could be that the 1st sample we heard was more heavily edited but then why wouldn't they edit it the same way here too?
But I can hear obvious drop in quality, it is important to note that my japanese language skills are limited but she does sound very muffled to me, beforehand I could make out the phonemes she's singing like:
I can make out that she sang "zenbu"
But here it straight up sounds like gibberish:
And one thing I am speculating on is that maybe since they're not pitching type A after rendering it but actually..the vb itself?? It's messing with the AI? Because here the tuning sounded a lot more messier, it was disgustingly pitchy and I could hear errors.
Last edited:
Holy crap, if that is true (and I've just noticed the drop in quality too after listening to those videos again), Kafu's bank is actually in a much worse final state than I thought. Yikes!
Please check out this post, I have some of the lyrics written out in higarana (it's actually "zenbu", not "tsunbu").
CeVIO - Kafu announced for CeVIO AI (Early 2021)
The details aren't public yet other than that it's coming in winter 2020/being a singing synth voice, but a virtual singer named Kafu (they write it as Kaf, though) (花譜) is collaborating with CeVIO. She started VTuber activities when she was only 13 years old (she is now 16) and her popularity...
vocaverse.network
Also, I'm dumb and realized the name of this demo is chick/baby bird in Japanese, so that's what the Japanese users were referring to.
I agree that the demo just sounds really pitchy/screechy at parts and the high pitched/nasallyness makes it hard to understand her phonemes. It just sounds like she's singing into a cup or something that's muffling everything.
I’m really glad I detached myself from her when her first sample came out... I can’t even imagine how disappointing this is to people who had hope that she’d turn out alright. I’ve never actively wished for a VB to fail, because I know taste is subjective and I’m happy for the people who like them... but I do NOT wanna see or hear Kafu regularly in the coming months. God help us.
To me it sounds like they used a raw, untuned render as opposed to a highly polished demo that isn't representative of the actual product straight out of the box, though I can't imagine why. It's definitely turned out to be a hype-killer. I still maintain the stance that any synth can sound good with enough effort, so I'll hear what other people have made before giving my final verdict.
Correct me if I'm wrong but isn't the appeal in AI synths that they come out ( mostly ) nicely tuned out of the box?
I realize this is product that is still in development but surely this can't be untuned due to the nature of AI synths and that raw vocal wouldn't necessarily be this pitchy.Yeah it's pretty natural to avoid criticising something directly, but the problem is there's no criticism at all... it seems the general commentariat can't even find any fault with it. Even my Japanese mutuals who are into voice synthesis have taken note of it and mentioned how it may mean she can appeal to a general audience but really it feels like this could just be Kaf's fans praising Kafu. She's absolutely not going to do well outside of Japan, and hell, I'd be surprised if many people are interested in her as soon as we hear Yukari or flower.
The most disappointing aspect is that, once again. This is a vocal synth that's probably only going to hit a very specific niche market. Even most people I know who are all for the all moe-high-pitched voice type aren't a fan of her voice. I'm really having trouble seeing KAFU hitting a market besides a very specific circle of people.
Yeah, for the most part you're correct and I wish it were true that they came without flaw, but with AI you're also getting the singer's flaws and weaknesses, along with their likeness. I really think they shouldn't have tuned Type A & B in the sample video, it quite clearly had pitch bends added and was pitch-corrected which made it sound far better, like... they've set themselves up. The mixing too, it put her at the forefront which made her sound better.
Unlike Saki AI's voice provider (who sounds like she's competent from the AI), KAF herself isn't technically a great singer so I can't imagine her actual voicebank is void of pitch errors and strange pitch bends because of her unusual way of singing.
Eh, I wouldn’t blame this on KAF herself. This is less due to her style of singing and more a clusterfuck of bad programming and horrid publicity.