VOCALOID Crypton new Miku NT ver.2, Miku V6 and Piapro Characters Super Pack

[blackout]

Yeah these definitely sound like the banks that they’ve been using in prsk.

I actually really like how Miku, Rin & Len sound, and I think I even prefer Kaito SP over his V3 Straight bank, but really conflicted on Luka and Meiko. They both have this really nasal sound to their vowels, but it really bothers me the most with Luka. Maybe I’ll warm up to it eventually but man… hopefully her V6 bank (assuming they’re not just doing Miku) will finally be the first bank since her V2 that I’ll actually enjoy right away.

Edit: Seems like Crypton’s panel at Magical Mirai for tomorrow, 31st August 3PM-4PM Tokyo time will be live streamed and archived in the Piapro YouTube channel. They will be showcasing future Piapro updates so… we could be hearing a snippet of Miku V6 and/or the NT update very soon!

 

[MagicalMiku]

talking session video at Magical Mirai 2024 in Tokyo (available from August 31st, time 15:00~)
Crypton staff will talk about Piapro Characters Super Pack, the new Miku NT and Miku Vocaloid 6(●'◡'●)

 

[Blue Of Mind]

Miku, Rin and Len sound like themselves, while SP Kaito is probably the best version of Kaito I've heard so far. Meiko and Luka sound the most 'off' compared to previous banks, however. SP Meiko seems to be a bit more nasal than even her V1 sometimes, while Luka sounds too cutesy compared to her V2 and even her V4, almost like she's been turned into a pink version of Miku. I'm guessing they may have used recordings for their NT banks for the SP versions, especially since a lot of people have complained about ProSeka songs using Luka NT in the past.

 

[wrong_thyme]

crypton: we didn't want ai bc we want the characters to sound like themselves!!
also crypton: *makes meiko and luka sound like miku appends*

i hope v6 is able to save them and kaito bc wow, they really don't know how to handle non-cutesy voicebanks do they!!!! im actually annoyed, they really managed to surgically remove everything that ppl actually liked about their voices and mikufied everything

 

[lIlI]

According to some folks, Luka's voicebank uses Miku's samples in certain areas - it definitely sounds like it, she's much closer to Miku than she used to be. Poor Luka fans, they haven't been able to catch a break since EVEC.

 

[BiscuitsandTriscuits]

My hopes have been shattered :(

 

[crtstatic]

I came across one of the posts about Luka SP on twitter earlier and...yeesh. That's not even Luka. They might as well have made a new character if that's the way they wanted to go about it. Like, especially when SP was compared to V4X. That's just not her.

 

[WyndReed]

I’m wondering if it’s not that they’re using Miku’s samples in Luka’s vb, but that there’s a way that they’re processing/formatting Luka’s samples which makes the vb end up that way. The super pack vocals definitely sound like the ports of the NT vbs used for Project Sekai. (Something that I’ve been mentally munching on for a while about NT is if they’re using some sort of AI to build the voicebanks and if maybe they trained it on Miku and then fed it data for the other characters. Like how the Synth Vs can all sound the same at times.)

edit: Does anyone have links to any samples of Meiko or Luka SP xsy’d with their older vbs?

 

[Midori]

That is NOT Luka

Once again, they did her so dirty...

 

[crtstatic]

Just figured I'd spit out my full thoughts before I forget:

-Miku, Rin/Len, and Kaito all sound fine, aside from how choppy the vbs are in general. Like, voice and tone wise they're fine.
-Luka sounds like her voice is phasing between a slightly pitched up version of her voice and Miku's. It's not pleasant at all to listen to.
-Meiko sounded off but not as much as Luka. I genuinely think Luka was the worst offender here.

I don't even like being negative about vsynth products, I try to keep an open mind...but I have to ask...who thought these were remotely ok to release???

 

[kimiworld]

Actually, I do not want to use the words "nasal" and "muffled" to describe their voicebank quality, because I do not think they truly sound like that. Based on what I have heard so far (especially the raw usages), the voicebanks sound more "solid" but at the same time more "stuffy", "tensed" and "thick" (I'm not sure if these words are even correct to use, but you get what I mean) compared to their V3 or V4X original voicebanks.

I can sense Crypton's intention to:

1. Standardize and uniformize the sound quality across different pitches, and
2. Bring the "voice colors/personalities" closer to their initial original V1 and V2 voicebanks.

I think this is what they meant by "by default should be unique to the characters' sound profile instead of becoming like their voicebank providers," because real-life vocals are dynamic across different pitches, and current AI singing voicebanks try to emulate the dynamic nature of human voices automatically rather than adopting hyper-realistic approaches by default.

However, through the released SP voicebanks, I found that Crypton seems to have overlooked:

1. The "clarity" or "crispness" of the vocals. This is important because, by default, the voicebanks should sound "relaxed" but not too "open-mouthed."
2. The fine-tuning of voice profiles to distinguish between each character, especially Meiko and Luka, where I clearly see Crypton's difficulty in handling the young-to-mature transitional (Luka) and mature (Meiko) female voicebanks, and
3. The fine adjustment needed for the transition between consonants and vowels. (This one is especially a big and important issue compared to the rest.)

Just for clarification, I use different labels for the voice qualities of the six Piapro characters, as stated below:

1. Hatsune Miku: young Japanese high school female voicebank
2. Kagamine Rin: child-like female voicebank
3. Kagamine Len: child-like male voicebank that is stuck at the beginning phase of male puberty voice change
4. Megurine Luka: transitional phase between a young Japanese high school female voicebank and a working adult mature female voicebank
5. Meiko: working adult mature female voicebank
6. Kaito: working adult mature male voicebank

 

[Mr_Meester]

Just figured I'd spit out my full thoughts before I forget:

-Miku, Rin/Len, and Kaito all sound fine, aside from how choppy the vbs are in general. Like, voice and tone wise they're fine.
-Luka sounds like her voice is phasing between a slightly pitched up version of her voice and Miku's. It's not pleasant at all to listen to.
-Meiko sounded off but not as much as Luka. I genuinely think Luka was the worst offender here.

I don't even like being negative about vsynth products, I try to keep an open mind...but I have to ask...who thought these were remotely ok to release???

Crypton has always been rushing their products so that they can fix them later. Rin and Len Act 1 were made in the same amount of time it took to make Miku V2 iirc. Miku NT is a good example of this. They pushed her out really fast and then pushed out an update that improved her quality. The Piapro Characters Super Pack feels like a rushed attempt to bring back interest into the Crypton Vocaloids.
I agree that I don't like being negative about this stuff, but I don't like what Crypton is doing.

 

[crtstatic]

The Piapro Characters Super Pack feels like a rushed attempt to bring back interest into the Crypton Vocaloids.
I agree that I don't like being negative about this stuff, but I don't like what Crypton is doing.

I made a comment to my friend earlier that it feels like Crypton knows they can just get money with little effort, because they know people will buy the voicebanks anyways. And with how rushed and unfinished these vbs sound...

 

[pico]

Hi friends,

I heard some buzzing secondhand that concerns have been floating around about Luka SP reusing audio data from Miku NT on their high pitches.

This had me concerned that something explicitly anti-consumer was going on here, so I decided to perform some frequency analysis to get to the bottom of whether or not this is actually the case.

The short thread I made on Twitter:
https://x.com/aptgetrobot/status/1829700132591284334

I've also attached the findings in PDF format.

TL;DR: there is no grand conspiracy, the samples are not the same. Luka SP does retain the nasal quality from Luka NT that made her unpopular when her use appeared in Project Sekai. So I don't blame anyone for disliking her, but wanted to assuage this one particular fear.

 

[kimiworld]

Hi friends,

I heard some buzzing secondhand that concerns have been floating around about Luka SP reusing audio data from Miku NT on their high pitches.

This had me concerned that something explicitly anti-consumer was going on here, so I decided to perform some frequency analysis to get to the bottom of whether or not this is actually the case.

The short thread I made on Twitter:
x.com

I've also attached the findings in PDF format.

TL;DR: there is no grand conspiracy, the samples are not the same. Luka SP does retain the nasal quality from Luka NT that made her unpopular when her use appeared in Project Sekai. So I don't blame anyone for disliking her, but wanted to assuage this one particular fear.

So does the analyses indicate anything regarding a claim about "the SP voicebanks were sampled from the so-called NT voicebanks"? The claim is currently spreading so widely on Bilibili which is the Chinese-equivalent Youtube. I think it is hard to determine for every single Piapro characters since only Miku NT has been released so far, so only comparison between Miku NT voicebank and Miku SP voicebank could be made.

 

[pico]

So does the analyses indicate anything regarding a claim about "the SP voicebanks were sampled from the so-called NT voicebanks"?

This analysis in particular would not reveal anything about that. We would need to perform a similar analysis on the NT data. If this is a point of concern, the best I might be able to provide is a comparison of frequency data between Miku SP and Miku NT samples. As you imply, because Luka NT is not a commercial product at this time, we have no way to perform this form of analysis with her.

My personal sense (not supplemented by data and analysis) is that they do sound uncannily similar so I understand where the bilibili users are coming from. If you have access to the SP products, I recommend taking a listen to Luka SP's raw samples and making your own determination.

 

[Nezuh]

I mean, I appreciate the frequency analysis and all, but at the end of the day Luka SP literally sounds like Miku.
She has the same distinct accent as Saki Fujita in certain sounds like [ i ] and some [ k ] phonemes.

Side note but from those same sounds I recognized Saki Fujita's voice in Shikanoko Nokotan lol

 

[lIlI]

Thanks Pico, that's really interesting!

This means that the samples must sound similar for one (or both) of these reasons:

1. Technical failure
Crypton is so used to working on Miku that they struggle to process samples from other styles of singing without Miku-ifying them. I don't have the technical expertise to speculate on how or why this could happen, or if it's even possible to do this accidentally; but it's a potential option.
OR
They couldn't get cross-synth to sound acceptable between drastically different timbres, so they homogenised everyone's sound. Something similar happened with Luka EVEC, which had less tonal variety than previous appends, ostensibly so they could be combined more smoothly.

2. Deliberate stylistic change
Miku is the most popular voicebank = we can make other voicebanks more popular by making them sound more like Miku ...or at least that's how I imagine Crypton's logic would go. While this initially sounds absurd, Vocaloid's primary Japanese demographic is men who prefer the youthful sound of anime highschoolers: 'cuter' higher pitched voices have been considered more desirable as far back as the creation of Sakine Meiko. It's not out of the question that Crypton would deliberately create 'Sakine' versions of their adult female voicebanks in an attempt at giving them more market appeal to COWARDS.

 

[Pup-p]

I like them Luka and Meiko in particular there is a bit of sounds in them that sound weird idk how to describe it… but overall I think this pack represents a part of the market that has been neglected for a while (non realistic vocals) and I’m glad that this release is happening!

It reminds me of the lagtrain pjsk cover. Especially the "sounding like miku" part (except its Meiko instead of Luka)

 

[MagicalMiku]

Crypton staff - talking session at Magical Mirai 2024 in Tokyo
8月31日（土）15:00～16:00

Piapro Characters Super Pack
-SP databases are originals and compatible with others previous versions V3 and V4X, so is possible to make different combinations.
-Super Pack is more about the charm of singing together, with harmony and fun.It's a good starting point.


-Growl and Cross Synthesis of Super Pack can blend those various expressions to make unique results in a song.


-The demo songs were made to demonstrate the bright and cute side of the Piapro Characters, so that's why they sounded similar and had almost a different signature from the past, but it's up to the producers to make their own original songs with very different voice signatures.

Miku NT ver.2
-Miku NT ver.2 early access will be available on October, with public release planned for first half of 2025.
-The early access version, also referred as beta version or pre-development version, will be free to download to owners of Miku NT.


-The whole engine M9 of Miku NT ver.2 has changed significantly, with original database and new features, such as Piapro Studio Automatic Control (to help customize settings as pitch and vibrato) and the accent function that allows for under-expression.
-New parameters for adjusting the singing have been newly developed for the NT2 engine.
-NT2 will be also optimized for lower-end PC hardware compared to now, so startup will be faster and there will be less use of resources.

-Development is going very well and Crypton staff is very satisfied with all the results. The following months will be for debugging and check any compatibility issues and hear all feedback from producers.
-The on-stage live demo of Miku NT2 was very impressive.

Vocaloid 6 and future products
-Regarding Miku Vocaloid 6, is currently under development together with Yamaha, with a planned release in late-2024.
-Vocaloid 6 can use the original japanese database to make songs in different languages, not only english.
-There is a alpha-version of Miku Vocaloid 6, and there are some challenges compared to V4X and NT2, so Crypton and Yamaha are working together to offer a good result. It's about giving users choices. Some things have to be tuned on Yamaha's side, so there is a lot of back and forth. There will be a follow-up report about Vocaloid 6.
-Crypton staff also hinted that Meiko, Kaito, Rin, Len and Luka will receive updates in terms of NT2 and Vocaloid 6,that's why they left some empty boxes in the line-up image of the presentation.
-Super Pack, V3, V4X, NT2 and Vocaloid 6 are all so different and can be used together to create unique and beautiful songs.

Hiroyuki Itoh (CEO of Crypton Future Media, Inc.)
Wataru Sasaki (Hatsune Miku producer of Crypton Future Media, Inc.)
Kiyomitsu Koizumi (Team manager of SONICWIRE music business team)