These are some raw samples extracted directly from LUKA_SP's ddb. I think this should settle things a bit.
VOCALOID Crypton new Miku NT ver.2, Miku V6 and Piapro Characters Super Pack
- Thread starter MagicalMiku
- Start date
These are some raw samples extracted directly from LUKA_SP's ddb. I think this should settle things a bit.
Dude this is going to drain my bank account
Why is Crypton releasing so much so fast??
NT2 demo at 38:25 for other people who are looking.
www.youtube.com
- YouTube
Enjoy the videos and music you love, upload original content, and share it all with friends, family, and the world on YouTube.
They mentioned that NT 2.0 will include autopitch tuning! This is similar to Maghni AI's original concept: AI driven tuning on top of a concatenative voicebank. I think this sounds like a good way to preserve the iconic robotic sound of Miku's original voicebank, while saving people some of the busywork of basic tuning.
Some questions I have for people who own the Super Pack:
- Is an internet connection ALWAYS required for usage? It says so in the requirements, but I'm not sure if it actually does need it.
- Will it work outside of Piapro? Like, will it work on V4, V5, and V6?
I've seen people use the SP banks in Tunelab and in V4 and V6, so yeah. Aren't they just V4 banks?
I decided to listen to something using Luka SP to see if the drama was warranted. Good God It literally sounds like she gets replaced with Miku whenever she hits a certain pitch threshold.
I just got it. I think Crypton is trying to make a “Great Value” version of the 6 voices lol
They are still selling the original voices after all, and this pack includes neither appends nor english voices.
Of course, in that case they should really advertise it as such, and it doesn't excuse the lousy quality, but at least it makes the whole thing kinda understandable.
And if they are really doing this, they are really hurting the image of their product lineup with this.
They are still selling the original voices after all, and this pack includes neither appends nor english voices.
Of course, in that case they should really advertise it as such, and it doesn't excuse the lousy quality, but at least it makes the whole thing kinda understandable.
And if they are really doing this, they are really hurting the image of their product lineup with this.
The original press release did say it was basically meant as a starter kit for people who don't already have a few of the voices, if I remember correctly. The V4X ones are the definitive versions. So I didn't have a lot of interest in the SP release to begin with, since it doesn't even support English, and I already have the ones with passable English support.
The NT demo is a lot more interesting. They've definitely made huge strides in quality, and I'm intrigued to see how it stacks up against the Vocaloid 6 voice bank. It may actually be competitive, especially depending on whether the V6 one retains its classic character or becomes more human sounding. I'm definitely more convinced that the new, as-of-yet-unreleased NT beta is what Deco27's been using lately.
The NT demo is a lot more interesting. They've definitely made huge strides in quality, and I'm intrigued to see how it stacks up against the Vocaloid 6 voice bank. It may actually be competitive, especially depending on whether the V6 one retains its classic character or becomes more human sounding. I'm definitely more convinced that the new, as-of-yet-unreleased NT beta is what Deco27's been using lately.
Vocals only of the Super Pack Singing Rolling Girl
I tried making my own Singer Settings for Meiko, Luka, and the others (not much changes to Rin/Len), and here they are singing. Personally I think I did okay for Meiko and Luka to sound good enough. I even tried improving my mixingprobably failed on it, sorry~ But I think I wished that Kaito would sound a bit better. He sounds a bit too rough, chopping(?) Robotic(?) when I tried different covers, but oh well. 
I kinda wish they didn't edit the samples of their voices too much to get them like this, but I guess I'll have to see for their v6 versions or NT.
A twitter thread to ease fear/angry of the Luka and Meiko are just Miku claims
A good read.
I tried making my own Singer Settings for Meiko, Luka, and the others (not much changes to Rin/Len), and here they are singing. Personally I think I did okay for Meiko and Luka to sound good enough. I even tried improving my mixing
I kinda wish they didn't edit the samples of their voices too much to get them like this, but I guess I'll have to see for their v6 versions or NT.A twitter thread to ease fear/angry of the Luka and Meiko are just Miku claims
A good read.
My inkling is that they are genuinely struggling with Luka on their technology. The sole reason for that being how precisely the issues users are complaining about with Luka SP mirror the repeated issues of Luka NT within the context of her Project Sekai demos. This is purely speculation, but I have a strong gut feeling that the SP vocals were heavily derived from work on the NT projects. I think they know what Luka is supposed to sound like and want her to fill the niche she used to given how Luka v4x went and how she continues to sound in her lower range. But for whatever reason, these pitches keep falling through since they initially split from the Vocaloid engine. It is very interesting.
Only tangentially related but I had someone ask that I share the analysis on the looping vowels, and the results are the same. Audibly miku-like but not the same samples with very different pitch behavior.
I got the vibe from Nezuh's post so just an explicit statement now that the intent of my research wasn't to convince people that Luka doesn't sound like Miku in this range (she absolutely does in my opinion), but rather to share that the databases are indeed different from each-other contrary to internet rumor. So while it's not as if Crypton is trying to sell you two different databases here when the data is actually shared between them, it is sad and unfortunate that Luka doesn't sound like herself and Luka SP would never be my personal first choice for a project where I need a voice like Luka's.
All that aside I was up watching the livestream and am incredibly excited about NT2, I think there are great things coming in the near future. And a warning to those anticipating the stated 2024 V6AI release that their phrasing was that they are "aiming, aiming, AIMING" for 2024. Hopefully they make the deadline. And if not it'll just be a little longer.
What are people's thoughts on NT2's eventual language opportunities?
To my mind, the reasons for the release of Miku on V6 AI is either
1) for people who want to use AI tuning, or
2) for its AI cross-lang capabilities (particularly for English, since it apparently benefits a lot from an AI engine)
If that's the case, I wonder if there will be any intention for an NT incarnation to support languages other than Japanese, or if those languages will be supported only by V6.
Given the choice, I'd rather get into NT at some point, but if it doesn't support English, there's not much reason for me to do that....
To my mind, the reasons for the release of Miku on V6 AI is either
1) for people who want to use AI tuning, or
2) for its AI cross-lang capabilities (particularly for English, since it apparently benefits a lot from an AI engine)
If that's the case, I wonder if there will be any intention for an NT incarnation to support languages other than Japanese, or if those languages will be supported only by V6.
Given the choice, I'd rather get into NT at some point, but if it doesn't support English, there's not much reason for me to do that....
For now I don’t think it’s in the books. They said in the magimirai presentation that V6AI Miku is filling the English niche for Hatsune Miku. So while it might happen eventually (and hopefully it does) right now they are positioning V6AI as the English product, not NT2.
I'm very happy with how English works in V6. Having one voice bank seamlessly handle both, simply based on the lyrics you type, is very nice. If they do add support for other languages to NT down the line, I hope they make sure the solution is fully baked and sounds good. I'm kind of assuming their priority will be updates to the other characters, whether on NT or V6 first, since they've been neglected for so long, before multi-language support.
It's also worth noting that V6 also supports Chinese, so there's another demographic being served by the same product. There is a Chinese V4X Miku already (which I've mostly heard in clips from Miku With You or that Genshin Impact event that featured Miku).
It's also worth noting that V6 also supports Chinese, so there's another demographic being served by the same product. There is a Chinese V4X Miku already (which I've mostly heard in clips from Miku With You or that Genshin Impact event that featured Miku).
While we don't have any official word on English for NT2, we do know that it supports English; as Luka NT English was used in ECHO. I would prefer English NT voicebanks over V6:AI, personally.
I really hope it materializes eventually. I think having both the cross-synthesized option and a concatenative NT bank would be extremely cool. I wonder though if Crypton will or does perceive concatenative technology as something embraced by the majority of the English-speaking community for their synths. Either way I sincerely hope it’s something we’ll see down the line.
My personal opinion I don't really care if they just stick to Japanese. japanese and english Miku just sounded so different from my ear, like a tonal shift from someone that knows two languages and sometimes feel like two different people. But if they could somehow make it consistant without relying on the inguinuity of Vocaloid producers sound design talent and make her sing bilingual, panlingual phrases seemlessly just with whatever they can bring up in just Miku V6 AI editor that would really be groundbreaking. Imagine Mongolian Miku to Khoisan Miku in single verses.
I'd personally prefer the idea of going to NT because
1) I feel like any new features that might come out are more likely there. (Though this is debatable. Yamaha may well introduce new features in a V7...which makes no difference, I guess, since it's not V6. And honestly, it seems likely that new NT features will probably take some time after release to work properly. I hate saying it, but that's the track record so far. I keep hoping that CFM will overturn that expectation.)
2) It might just be a case of me not having gotten used to it yet, but I always found the V5 UI rather unintuitive, with singers being selected and parameters changed in odd places/buried under menus. And, AFAIK, V6 still basically has the same UI as V5. I'd love to hear what experiences anyone has had with the V6 UI, though.
3) The fact that vocal presets rely on onboard plugins that can't be controlled from the DAW at all/integrated with your own effects chains doesn't help, either.
1) I feel like any new features that might come out are more likely there. (Though this is debatable. Yamaha may well introduce new features in a V7...which makes no difference, I guess, since it's not V6. And honestly, it seems likely that new NT features will probably take some time after release to work properly. I hate saying it, but that's the track record so far. I keep hoping that CFM will overturn that expectation.)
2) It might just be a case of me not having gotten used to it yet, but I always found the V5 UI rather unintuitive, with singers being selected and parameters changed in odd places/buried under menus. And, AFAIK, V6 still basically has the same UI as V5. I'd love to hear what experiences anyone has had with the V6 UI, though.
3) The fact that vocal presets rely on onboard plugins that can't be controlled from the DAW at all/integrated with your own effects chains doesn't help, either.
Miku e
2) V6, at least for AI voicebanks, has less tabs overall. However, it's still very unintuitive to work with and just ugly.
I would prefer NT because its UI/UX is just generally better in all aspects, and it wouldn't have to deal with the weird issues that happen in V6:AI outside of the sound quality. Yeah her accent would be strong, but I'd rather have an accent and phonemes that work instead of V6's poor cross-lang model where most vocals can't pronounce a lot of core sounds (most notable ar/er/ir/or).
1) VocaPigeon recently went to the Yamaha booth at Magical Mirai and they said they are planning on adding things over time to both V6 and VX-B, so that is something hopefuly. However, it does seem like NT2 is still Crypton's primary focus.
2) V6, at least for AI voicebanks, has less tabs overall. However, it's still very unintuitive to work with and just ugly.
I would prefer NT because its UI/UX is just generally better in all aspects, and it wouldn't have to deal with the weird issues that happen in V6:AI outside of the sound quality. Yeah her accent would be strong, but I'd rather have an accent and phonemes that work instead of V6's poor cross-lang model where most vocals can't pronounce a lot of core sounds (most notable ar/er/ir/or).
If those issues arise, how do you fix them? Add an extra -r phoneme or something?