auspicious-voice
AUSPICIOUS VOICE
This is the official voicebank distribution site of virtual singers Fuwa Maria and Fuwa Mario, maintained by HIRATELIER.
375 posts
Don't wanna be here? Send us removal request.
Last Seen Blogs
princelestatdelioncourt
Prince Lestat De Lioncourt
unkind-raven
Unkind
construction-of-reality
An Enemy of Fate
bleatbrew
BleatBrew
zeykad
ZeyKad
Text
I kept slacking off on this, but I made new logos for Maria and Mario!! I'm so happy with how they turned out this time, and I also updated their voicebank downloads with them :D
1 note
·
View note
Note
Hi , I really like the Fuwa twins! I have some questions them.
-if one is crying, what will the other do? Which twin cries easier?
-since they are “half”-qilin, can they catch a cold or other kinds of human sickness?
-who would win if they got into a fight (pyromancy allowed)
-what is their sleeping schedule like? Are they night people or morning people? Do they sleep for a long time or take short naps?
Hi anon! I'm always happy to answer your questions~
If Maria was crying, Mario would try his best to console her but also to remain respectful of her boundaries if she really needs some time alone. Now if it was Mario crying, then Maria would try to cheer him up by bringing him either Popo or Crumbie (their pet cats, I haven't drawn them yet). She doesn't know how to properly deal with people who are grieving, but she's just trying her best to work on that.
Yup! The twins can get sick like any other human would. The difference is that they can recover faster, but that depends on how severe the sickness is. For a frame of reference, if one of the twins caught a bad fever, they can recover from it in as little as 4-6 hours and then carry on as if nothing happened. If it's just a cold, then it'll take under an hour for a full recovery.
OKAY SO. Both the twins have their own expertise in fighting, with Maria specializing in quick-draw swordsfighting (think Vergil from Devil May Cry) and Mario specializing in hand-to-hand combat. When it comes to skill, Mario would probably win in that department as he's had plenty of experience seeing through Maria's attack patterns and disabling her weapon. In terms of pure strength however, Maria would win thanks to her destructive pyromancy coupled with her deadly swordmanship.
I like to think that Mario tries to maintain a normal sleep schedule - from waking up in the morning to going to sleep at night. However, he does like to take short naps throughout the day if he can find the time for it. Maria tends to wake up late morning or afternoon and will go to bed way after midnight (it's rare if she does wake up early in the morning). Unlike Mario, she doesn't like taking naps but she does find herself almost dozing off at times. She has a bit of a messed up sleep schedule as a result, but she can't be bothered to fix it.
0 notes
Text
youtube
Mario's DiffSinger demo is here! No I go work on the demo reel...
0 notes
Text
youtube
The official demo for Fuwa Maria's Diffsinger voicebank has been released. Please give it a listen!
1 note
·
View note
Note
eats the fuwas cutely
*you are immediately lit ablaze cutely*
2 notes
·
View notes
Text
Fuwa Maria AI & Fuwa Mario AI for DiffSinger Progress Report (April 2024)
It's been (almost) a month since my last progress report, but since it's April, I thought it wouldn't hurt to update with a new one. Plus, I got some big news to share!
Everything else is under the cut as usual.
Voicebank Progress
Both of Maria and Mario's DiffSinger voicebanks have been trained up to 200k variance and 160k acoustic! At that point, both voicebanks sound their very best, at least in version 1.0.0. Regardless, I'm still happy with the way they turned out, and at least they sound distinct from each other.
The reason why I say version 1.0.0 is because I have a potential version 2.0.0 update in the works. The thing is, version 1.0.0 will be the last time my old recording setup was used to record singing data. I say this because my current microphone, the Blue Snowball iCE, is just not cutting it for my needs these days, especially when it comes to singing in general. I don't really know when version 2.0.0 will come out, but maybe next year or so - IF I can get my hands on new equipment. Please expect version 1.0.0 to come out later this year, probably during June or something like that.
In the meantime, once I finish the artwork of my DiffSinger designs, I will post silhouettes of them along with new short demos. If that ReFlow update for DiffSinger comes out after I release my voicebanks, there COULD be a 1.1.0 update.
Possible Audio Equipment Upgrades
So for audio equipment, I'd like to upgrade to using an XLR microphone and interface for recording. I'm glad there are cheaper options out there that are actually good, plus I am on a budget myself, and would like to spend under $200 for that.
I am still using my pop filter, desktop stand, and isolation shield from my old setup, so with that being said, I'd like to upgrade to a Mackie EM-91C microphone and an M-Audio M-Track Solo audio interface. Both of these are budget options, but based on the reviews, I've heard that they are solid choices for starting out with an XLR setup. Plus the Mackie EM-91C already comes with a shockmount AND an XLR cable!
Potential Megamodel Development
I am considering developing a megamodel with my friends to add more language compatibility and range with the 2.0.0 update - that is, if everything goes to plan. I plan to add Maria and Mario's potential new datasets to the mix of course, as well as my friends if they're fine with that. It could be just plain singing data or even UTAU recordings - anything helps to further improve the model output.
For language support, Japanese and English are planned, but if possible, there could be support for French, Spanish, Italian, or perhaps even Chinese, Korean, and/or Tagalog. I still need to figure out the phonemes and how I want to organize them though.
I just hope that when the megamodel dataset is complete, Colab will still be merciful to me...I'm not sure why it takes longer to set everything up...
Character Progress
Designs
I just finished working on Maria and Mario's reference sheets for their DiffSinger designs! Currently focused on finishing their fullbody artwork, which won't be revealed up until their voicebank release.
And no, I don't want to work on additional append designs, that's already too much work...It'd make sense for standalone appends, which is what I did for my other UTAU voicebanks, but for DiffSinger...naaaaah.
Profiles
Nothing new at the moment! I'm still working on fleshing out the lore, given that DiffSinger Maria and Mario are just slightly older versions of their UTAU selves. Expect some lore drops about them when their voicebanks are released? I don't really want to make separate profile pages for them at all.
Anything Else?
For the potential 2.0.0 update, I would love to revive a discontinued UTAU of mine and resurrect them as a DiffSinger through the planned megamodel with a serviceable amount of singing data and vocal modes. If I can do the voice again, I might go forward with it!
Also Maria and Mario's 9th anniversary was on the 4th, and it's wild how they've existed for so long. Listening to their old voicebanks reminds me of how far they've come, and it makes me want to tear up a bit.
Anyways, I'll see you guys around! I might be able to publish full-length demos of Maria and Mario's DiffSinger voicebanks on YouTube at some point, so keep an eye out for that.
0 notes
Text
Cover art I made for my UTAUs' 9th anniversary! Check it out here.
2 notes
·
View notes
Note
Hello :)
I like fuwa twins very much! They are so cute! Btw, I have two question for them.
1) Is Mario still having jagged teeth? I found it in mmd model but I could find info about the twins' teeth only in UNUSED file... Can I check their teeth??😇
2) Which first person pronoun do they use if they speak in Japanese?
It's ok to ignore questions (or answer just one of them) if you don't want to answer. I'm not good at English, so I'm sorry if I sound rude.
Hi hi! I'm so happy to hear that you like them :D It's been a while since I got a new question in my inbox, so let's get cracking!
The jagged teeth is actually a detail I forgot to implement in Maria and Mario's updated artwork and reference sheets (since January 2024), since I barely draw them with their teeth details shown 😭😭😭. It was admittedly a cool detail I implemented in their designs before, but I guess it's no longer a canon detail in their UTAU designs, so they're just back to having regular teeth. And speaking of their MMD models, they really need a huge overhaul huh...they were made using the old reference sheets I had in hand.
I haven't actually thought about this in a hot while :0 Had to do some research on Japanese pronouns beforehand! So I like to think that Maria would use "watashi/私" for first-person pronouns, and Mario would use "boku/僕" and "watashi/私" for first-person pronouns interchangeably.
Hope these answers will help, anon! I always love receiving questions like these as it helps me flesh out Maria and Mario's character profiles a bit more.
0 notes
Text
youtube
Happy 9th anniversary to Fuwa Maria and Fuwa Mario!
Been working on this on-and-off for a while but then my life got a bit busy so I wasn't able to make a full animation for it...Hopefully the simple art and video I made will compensate. I'll probably relax for the rest of the day and binge watch some Vinesauce LOL
2 notes
·
View notes
Text
After weeks of training, here's a sample of Mario's DiffSinger voicebank (finally)! I'm happy with how it turned out, and I hope he sounds even better with the finetuned vocoder I prepared for him...
Also he sounds so good in the lower ranges, which will eventually be showcased in his release + demo reel! Now that the whole voicebank development part is finally dealt with, I can get to working on the reference sheets and then the fullbodies, and then the demo reel itself...I have a lot to juggle with and I haven't even finished the animation for Maria and Mario's 9th anniversary next month...help
2 notes
·
View notes
Text
Here's a new sample of Maria's DiffSinger voicebank! Unfortunately the voicebank used in this one is outdated as I trained it up to 200K variance and 160K acoustic for better quality, not to mention that I recently finetuned a vocoder just for her...
2 notes
·
View notes
Text
Fuwa Maria AI & Fuwa Mario AI for DiffSinger Progress Report (March 2024)
As of last month, I have dedicated my spare time to developing DiffSinger voicebanks, and it's safe to say that I've made a lot of progress in doing so. Also Maria and Mario's 9th anniversary is fast approaching, so I think it's about time I get around to making the art for it.
I think at this point, I may get around to releasing my DiffSinger voicebanks at around the 2nd to 3rd quarters of this year, depending on if everything goes smoothly as planned - and I have a demo reel and release video planned. But as usual, the bulk of this report will be under the cut!
Voicebank Progress
Fuwa Maria AI
So far, Maria's DiffSinger voicebank has been trained up to 108k steps variance and 134k steps acoustic! However, I intend to train it further to 125k steps variance and 150k acoustic to ensure more quality output, provided that it doesn't start to overtrain before that.
She sounds pretty good for the most part! I figured that at this point, her Spectre vocal mode is more of a calm singing voice, and that her Lapse vocal mode is more of a lively singing voice - but their expressiveness can be increased with the tension parameter, which I tend to abuse a lot.
I do have a sample of Maria's earlier DiffSinger build which the acoustic model was trained at 82k, but it is a complete improvement over her previous build from last month. Once I further train her voicebank, I will post another sample with a teaser of her new design!
Fuwa Mario AI
I managed to train Mario's variance model to 108k, but like with Maria, I will train it even further. As of writing this post, his acoustic model has been trained up to 30k steps, and I'll probably end up using all of my Google accounts to keep training given Colab is stingy with free resources and what not.
I don't have much to say for him yet as I don't have a workable voicebank to use, but expect a sample in the next couple of weeks or so! I just hope he sounds different from Maria...
Character Progress
Designs
I went through many design concepts but I think I managed to come up with something for once 😭 I can't say anything just yet BUT I am happy with how it's turning out.
There are still a bit of artworks I want to finish up (including a speedpaint I've yet to upload and the 9th anniversary art) before I start working on the fullbodies.
Profiles
Still have a vague idea for Maria and Mario's character profiles, but the gist is that they're older versions of their UTAU counterparts, and I'd imagine that their mental/physical age would be at least 22 years old or something. I guess that's that...for now.
1 note
·
View note
Text
youtube
Recovering from cover burnout so I made this!
4 notes
·
View notes
Text
Surprise! A quick test of Maria's DiffSinger beta, trained at 42k acoustic and 80k variance!
She's underbaked just a bit since this is a beta voicebank after all, but I am loving with how she sounds as an AI voicebank. Eventually her final build will sound not as scuffed and have more features implemented. Her vocal modes sound pretty alright so far, but hopefully they'll be more pronounced once I make the final build.
That being said, I'll be going back to labeling once again... 💀💀💀
6 notes
·
View notes
Text
Fuwa Maria AI & Fuwa Mario AI for DiffSinger Progress Report (February 2024)
It's been a while since I made a post like this, but I figured it's time to make another post for DiffSinger development shenanigans. I haven't uploaded a full cover in almost two months and I'm slowly working on Maria and Mario's 8th anniversary project alongside my other OC projects...which have been receiving lots of tending to thanks to being hit with new ideas and concepts. January was a productive month for me, so there's much to be discussed about what's going on with current DiffSinger voicebank development.
As usual, the bulk of this report will be under the cut!
Voicebank Progress
Still, I do not have any audio demos just yet. But the development is coming along well so there's that? I haven't even done the training part...
Fuwa Maria AI
I finally labeled Maria's ENTIRE dataset nearly after a month's worth of labeling! I had to deal with the pain of listening to my own voice for hours on end, so I'm hoping once I get to the training part, Maria will at least sound decent.
The only things left to do are to use SlurCutter on the dataset, but I'm having major troubles with especially when it comes to installing parselmouth on my Windows PC...I'm still new to using Python...but I am contemplating on doing all the SlurCutter work on my Linux PC if all else fails, otherwise setting up everything before autopitch training is plain frustrating 😭😭😭 If all else fails I might not train autopitch w/ SlurCutter at all...so in that case, hopefully my singing will at least suffice. Looking forward to vocoder training, however, so I might fine-tune the new HiFiPLN vocoder that came out recently!
I'm still holding off from training the dataset, as I am still waiting for the tension parameter to be implemented into the main branch...But once that gets done, I am also implementing parallel training for additional language support in the stable version! English is among one of them, and I'll probably throw in a Japanese dataset for improved pronunciation or w/e, and Chinese and Korean support would be nice too, but IDK if I'll ever use those languages LMAO...but perhaps I might train a beta version w/o tension and parallel learning soon? Let's see if Colab will be kind to me haha, and I don't have a PC powerful enough for local training...
Update: I in fact did train a beta voicebank for Maria! you can listen to a short snippet of what she sounds like here. Since the tension param won't be implemented for a while, I thought it'd be best to at least train a more stable version of Maria's DiffSinger voicebank with English crosslang and autopitch for public release, and perhaps work on a new version with tension implemented...but also I saw on DiffSinger's to-do board with rap and tone support AND attack and release...I'm excited for whatever DiffSinger has in store for new features :>
Fuwa Mario AI
I finished recording Mario's entire dataset (normal, power, and soft modes), and as of writing this post, I just started labeling it. He's more tolerable to listen to compared to Maria, but he's my natural voice anyways. I'm on my 4th song, and labeling isn't that much of a tedious process anymore...
Not much I can say, but expect him to also receive additional English support and such! Let's hope he'll sound good after training! I'm also hoping to finish labeling around the time tension is properly supported, but I don't really know when that will come out. Hopefully before their 9th anniversary arrives...
Character Progress
Designs
I STILL don't have a solid character concept for Maria and Mario, and that includes their design. Well at least I know what their hairstyle will look like...
I do have some concepts in hand but I don't know where I put them...
Profiles
I think I'll just stick to Maria and Mario's current character information...I just don't want to make up new lore for them as they've already gone through so many revamps.
Signing Off
I'm still into OCs as usual! But with new DiffSinger voicebanks releasing coming out, I'm more inspired to finish my own, so it feels like I'm in my UTAU noob days from 2014...and HOLY SHIT it's been 10 years since I got into UTAU...
I'm still treating vocal synths as a hobby, but sometimes it's nice to check up on what the folks are up to these days. Plus I got a new gunpla model kit and it's so pretty ^^ I think I built three so far!
See you around,
-- HIRATELIER
0 notes
Text
It took me weeks upon weeks, but I FINALLY finished new fullbody artworks for my UTAUs Fuwa Maria and Fuwa Mario! I also updated their voicebanks with the new art and updated reference sheets ^^
For those who have been following me for a while and have little to no idea what UTAU is, it's a free-to-use vocal synthesis program created by Ameya and it functions similarly to VOCALOID, but you can make your virtual singer of sorts. Personally I haven't used UTAU in a while because of burnout, but it's nice making things for my virtual children other than just making them sing LOL
5 notes
·
View notes
Text
Loved the idea of Mario growing out his hair, so I made changes to his character design to reflect on that as I work on his fullbody artwork! He's gotten kinda tired of keeping his hair short, so he's letting it grow long for change but not too much...
The speedpaint video can be found here! I made it a goal of mine to upload speedpaints to my YouTube channel whenever I can :D They're fun to make
5 notes
·
View notes