Other Maghni AI by VocaTone Studio and Misbah Studios

[lIlI]

March progress report
Notable events: First engine synthesis (not yet demo ready)

Launch libraries have solid numbers, seven languages planned for launch:

I'm not sure if '6 libraries' refers to Aurum and Audine's combined six voicebanks (three languages each), or six separate characters.

Okay, Kiyoteru taught me maths and I figured out that it has to mean six vocalists - including Aurum & Audine - covering seven languages between them, with English, Japanese, and Spanish confirmed.

 

[Silver1063]

Well, seeing as we already have English, Spanish, and Japanese voices confirmed, and counting Aurum and Audine as 6 libraries wouldn't add up to 7 languages, I'm guessing there is planned to be at least 4 more unique singers to get to 6 "libraries", with some combination of 4 different languages for them. I'm thinking Chinese, Korean, or French would definitely be there. Maybe Oliver would have Chinese for some reason....

 

[lIlI]

Yup haha, I just realised as I reread it.

Time to guess wildly about the four mystery maghnites:

• Sonika update (English only)
• I agree that Oliver (English/Chinese) sounds likely
• The deep Jazz vocalist Natalia mentioned (Arabic/French/English)
• A rock focused female voicebank (English/German) [shameless wish fulfilment] or the Brazilian jazz singer mentioned (English/Spanish/Portuguese)

I think the latter two would be Misbah Studio's flagship voicebanks aimed at non-fandom musicians, with perhaps a photo of the singers instead of stylised characters, while Sonika and Oliver would be VocaTone's fandom push.

 

[___]

Maybe it could be one language that is phonetically ( and in other aspects too ) so close and near identical to another thus allowing you ( + with the help of the promised fancy features??? ) to use it for that language as well w/o problem?

 

[___]

Thank you @Silver1063 for asking, I was skimming through the replies and noticed this so I hope it's ok to post here!


I don't wanna be negative Nancy so all I'm gonna say is that I hope they've really made enough progress behind the scenes to make such statement justifiable cuz....6...is a lot...and here I thought making whole new synth from scratch and 2 vbs capable of 3 languages was a lot of work for only half a year...but wow if this didn't come out of nowhere

The thing is despite some major slip ups in the beginning and despite them claiming they wanna be very open about the whole process I have caught on that they're extremely sneaky and careful with their wording.....which makes me not able to make any definitive claims....we'll have to see....and besides I don't wanna end up eating my words

ok I think I failed at not being negative Nancy orz

edit: also double post orz

 

[morrysillusion]

ive always been a bit unsure of their wording on many things, and it makes me wonder a lot if what we assume is done is not as done as it is. even their first and second Q&A had a lot of confusing questions about their progress came up, often wording that i think again made people think something was in fact done when it wasn't. like about demos, the engine being done, etc....

i do appreciate the fact they didnt jump on that kickstarter right away, but i guess i cannot help me be skeptical ever since that first Q&A that had too many back and forths about what the situation was about many parts of it.

 

[peaches2217]

This is my main issue with the development team in full view: they don’t even have a functioning, demo-ready engine, and they’re already talking about their six starter languages and 44 eventual languages and— like, I would adore an engine that could do all that and do it well. But they’re so heavily focused on what it can be that they’re jumping straight to that, before they even have a demo-ready engine?

An engine that can hypothetically support 44 languages but is only 10% completed after several months is no better than an engine that can only support one but can at least function. I just wish they’d focus as much attention on getting a solid launch product out as they do on daydreaming up the next seven obscure languages they’ll feature. It’s very frustrating to watch.

 

[morrysillusion]

This is my main issue with the development team in full view: they don’t even have a functioning, demo-ready engine, and they’re already talking about their six starter languages and 44 eventual languages and— like, I would adore an engine that could do all that and do it well. But they’re so heavily focused on what it can be that they’re jumping straight to that, before they even have a demo-ready engine?

An engine that can hypothetically support 44 languages but is only 10% completed after several months is no better than an engine that can only support one but can at least function. I just wish they’d focus as much attention on getting a solid launch product out as they do on daydreaming up the next seven obscure languages they’ll feature. It’s very frustrating to watch.

the language focus is 100% the stuff i have had an issue with....as far as im aware as i think they clarified after many questions, it seems "implemented language" is just...having a recording list written? (which like...anyone could do with or without an engine) in the end that has nothing to do with voicebanks. it just says "yeah, you could record it if you wanted to" not "we have recorded tests banks of all these languages and know they will work". the language stuff has everything to do with interest. writing up 44 rec lists means nothing at all, imo. i hate for that to sound so negative but i think people who are following them get way too hyped for what they think it means, and this team keeps focusing on it. i could care less if they really do record six language banks! i care way more about hearing if they have a engine and functional UI prepared.

 

[___]

i do appreciate the fact they didnt jump on that kickstarter right away, but i guess i cannot help me be skeptical ever since that first Q&A that had too many back and forths about what the situation was about many parts of it.

ngl I always thought that the community having to convince them into doing so was pretty bad in on itself...just didn't wanna be overly negative about what ultimately was a right decision :/

as far as im aware as i think they clarified after many questions, it seems "implemented language" is just...having a recording list written?

i hate for that to sound so negative but i think people who are following them get way too hyped for what they think it means, and this team keeps focusing on it.

Yeah you're spot on, that's what exactly meant by them being sneaky with words.

The heavy language focus' been explained and I understand there's only so much they can do with the limited team they have but at the same time I can't help but feel like """filling in the time""" while they wait for their single coder by writing 40 language reclists is just poor management

 

[peaches2217]

Reading back through it... six VBs at launch, all of who will be multilingual... that's, at the very least, fourteen launch VBs (the first two's three languages and the other four supporting two), 21 if they plan on all of them having three. Even if their Kickstarter gets funded, their team is small, and their engine will be in infancy. How on earth do they plan on assuring quality control on all of those? There's a difference between vocals who can sing in multiple languages and vocals who can sing well in multiple languages. Even V5, with all the backing and funding of Yamaha, didn't try pulling that off; with a team that small, doing so without it being an absolute disaster would take years.

If I'm reading correctly, they've just finished synthesizing their first English vocal, four months after starting. Now let's give them the benefit of the doubt and say it's a funding issue, and they could get VBs completed faster if the Kickstarter is funded. Still, if they plan to do all of those VBs in-house, it would take a minimum of two years with their current team. And that's just their launch vocals, to say nothing of the various additional languages, vocals, collabs, etc. that they have planned after launch.

Though it wouldn't surprise me if they haven't thought that far ahead, heh.

 

[CdrSonan]

I find it extremely hard to estimate at which stage their development is based on what they‘re saying. Perhaps I‘m overanalysing it, but a lot of their wording can be interpreted in two ways, where they‘re either almost done, or have hardly anything.
About this massive amount of voicebanks, this is pure speculation, but perhaps they‘re planning to record them at the same time as they‘re programming a lot of the engine. It would explain how they plan to hit such a tight schedule, but it would also mean that the voicebanks will be tested with an unfinished engine version. I guess with enough knowledge about phonetics (which they definitely focus on a lot) you can pull that off, but it still sounds like a quality control nightmare.
Or maybe they do only record the vocals after finishing the engine, and development is a lot farther than it seems

 

[pico]

I'm glad this team seems to be receptive and at least mostly communicative. Putting off the Kickstarter was 100% a smart move, and frankly, I question why it was even on the table. This project has all of the hallmarks of past Kickstarter campaigns gone wrong: no tangible product demos from project announcement, immense promises that strongly resemble feature creep, and 'extra' marketable work done before the product itself, i.e. character designs. (Yes, I understand that a lot of the team is sitting on dead time, but it doesn't negate the trend.) Not to imply a project can't be successful with these characteristics- they definitely can be. It wouldn't have been a good look, though.

Another thing that sorta raises my eyebrow is the apparent one programmer. Like, sure, we've seen single-developer voice synths in the past in UTAU and I think NEUTRINO, but the difference there is that NEUTRINO is a straight neural net (which this engine is not) and UTAU was in development for at least 3 years before supporting triphones. (In fairness, Ameya was developing UTAU as a hobby where this programmer has what I assume is a full-time job, but I digress.) I just hope a single-person team is enough to develop the high-quality systems they promise. Man, just... good luck, programmer. I'm rooting for you.

 

[Leon]

Wasn't Kanru a one man team for SV's initial release? It can be done, with the right care, knowledge, and time. Here's hoping the MAI team has that! I am cautiously optimistic. It doesn't even have to be some top tier A+++ product right out the gate, to me. Decent voicebanks with user friendly tuning and sensible phonetic editing are what I'd like. Also preferably a better engine noise sitch than CeVIO,

I'm rooting for you, MAI!!

 

[___]

I think we all here root for MAI to turn out well but it's important to point out it took Kanru 3 years to make the engine alone ( 5 if you count moresampler as the resampling SynthV engine def has roots in moresampler ) and on launch ( counting only in-house solo vbs for fair comparison ) it had two incomplete vbs ( Ellie and Genbu and I love Genbu but he really was rough around the edges on release ) and an UTAU port and supported 3 languages. Vocatone and Misbah's goals are larger scale than that and that's ultimately what's cauzing the skepticism :/

ah sorry to be downer, like no one here wants it to fail ofc.

 

[Prism]

I would like to chime in here the amount of banks might be for testing and never see a commercial version. Kanru has said in the past that there are way more voice banks made than the ones that are commerically available. I know even vocaloid had spanish voice banks in v1

 

[peaches2217]

I would like to chime in here the amount of banks might be for testing and never see a commercial version. Kanru has said in the past that there are way more voice banks made than the ones that are commerically available. I know even vocaloid had spanish voice banks in v1

They explicitly said they were gonna have all of them at launch though, so they're clearly meant for commercial release. You bring up a good point though: if they were made for testing and working out the kinks rather than for release, and maybe released after they had a stable engine, that would make a lot more sense.

 

[Prism]

I wonder if they have even recorded any of the banks yet you don't really need a program finished if you have a plan for how the reclist works. It seems like they are passionate about it so they are probably making personal banks a releasing it. But it does seem like scope creep.

 

[TheStarPalace]

April progress report was posted, they seem to have worked on the engine a lot and have three beta voicebanks called ENG+M1, SPA+M1 and JPN+M1.

 

[frankensalad]

Sounds like they're finally getting some of the groundwork down, which is good. I want to believe in this project.

 

[lIlI]

In a reply to their latest monthly update (see OP for the list), Misbah Studios indicated that Audine's demo may come within the next few weeks: