Other Crowdfunding for A.I.VOICE Adachi Rei, TTS without a voice provider

[uncreepy]

A Twitter user called missile_39 has begun crowdfunding on CAMPFIRE hoping to produce an A.I.VOICE text to speech voice bank for Adachi Rei (足立レイ).


Her special trait is that she is TTS without a voice provider (timestamped):

Her voice is produced through overlapping Sin waves. The method for her speaking is called Replivoice (レプリボイス), being a reference to her voice source being a "replica of synthetic sound". If the crowdfunding succeeds, the product will be named Replivoice.


Rei's UTAU speaking:

There are some parts that sound screechy/feedbacky and missile_39 acknowledges that UTAU is not really meant for speaking. Her voice is supposed to sound synthetic yet feminine. You can download her UTAU here.


Rei's UTAU singing:


She is honestly somewhat hard to understand. In addition, she is supposed to be a life-sized bishoujo robot.

missile_39 has been working on the robot version of her and there are several parts so far:



2 years ago, missile_39 contacted AI, Inc. to ask if it is possible to make a voice bank with a synthetic voice instead of a human voice. Normally, the voice provider must read 500 sentences and have their voice split into phonemes. But Rei has no voice actor, so nothing can be recorded. Apparently AI, Inc. is on board with trying to work together to do this backwards/slightly strange version of TTS.

Rewards for crowdfunding include a package of Re's A.I.VOICE bank, a 3D model with a unique chest number on a USB (the model is made by missile_39). There is a version of the 3D model labeled as #00 already being distributed on nico nico. The format is for VRM and MMD. Supporters get their name put in an announcement video.

Package:

Model #00:


Custom chest number:


Note: This post is getting kind of long, I will continue with more information in my next post.

 

[uncreepy]

Tiers in the crowdfund help fund each necessary 3D-printed resin body part for her (each part is almost $1,000).


I downloaded the PDF of their Miku robot here:


Stretch goals will fund happy, sad, and angry voices. But each emotion costs adds more to the combined cost. Normal voice = $30,000. Adding Happy = $40,000. Adding sad = $50,000. Adding angry = $60,000. They want to add different robot face emotions to correspond to each emotion that is funded.

Delivery of funded voice(s) will begin in September (will take about 10 months).

missile_39 is currently attending college for mechanical engineering and is researching bipedal walking robots + synthetic speech.

For Q&A, the cost does not include recording cost, because there is none due to being synthetic speech. If the crowdfunding succeeds, they will need to make a small corporation for sales (which will be sold through mail delivery only) due to AI, Inc.'s agreement and they need to keep the sales small scale.

 

[Leon]

I was just reading your thread on this on Twitter. It's some really interesting stuff!! I like the mini explanations on how her voice works.

I have no need for a speech synth, so I won't be buying her or anything, but it's fun to see these projects and what people do with them! I hope she gets funded.

 

[uncreepy]

I found footage of robot Miku and she is bigger than I imagined, but she moves when the person moving moves their body:


Making of Miku video:


Miku can walk when held up by ropes (timestamped):

I realized that she is probably named Rei because that's one way to pronounce "zero" in Japanese. This video around 2:45 shows her with the 00 numbering. missile_39 appears to be a big Miku fan, so I guess this might kind of be a Crypton insert character (especially since Miku appears with Rei in a sort of mini comic).
Info around 3:00 says she has wi-fi and bluetooth antenna by her ponytail. She also has a servo mechanism on the other side of her head.
Thighs have electronic patterns. Her sneakers say "adachi" on the tongues.
Her clothes are supposed to be padding for when she falls down or gets bumped into to protect her robot parts.

 

[Blue Of Mind]

Definitely an interesting concept! I also quite like her character design - it kind of reminds me of 2000s-era anime-styled games set in the future. Her UTAU voice reminds me of a distorted Teto right now, but I assume she would sound very different and less distorted in A.I.VOICE. (As handy as UTAU is for potential projects, I find it doesn't really demonstrate how a final VB will sound like at all, because UTAU's engine processes voices differently in comparison to whatever software a finished VB ends up on).

 

[ikaroll]

Oh my god I love her voice! Her design is really cool too!

I’m interested to see how she sounds in A.I Voice!

 

[pico]

I love Adachi Rei and robotics. I totally get missile39's passion for this project and I hope the crowdfunding follows through!

 

[uncreepy]

AI, Inc promoted the crowdfund on their A.I.VOICE page:

 

[poser]

i love adachi :) i remember finding out about her a year or so ago. really hope this goes through, shes awezum

 

[uncreepy]

A lot more people suddenly donated today. So now there have been 29 donators and it jumped up to collecting about $8,000, 25% of their $32,000 goal. There are 17 days left, so I feel like she probably will get funded after all at the rate it's gaining traction. Overall, I think people are impressed with the novelty of her voice and people generally seemed to react positively to Rei (in both English and Japanese).

 

[uncreepy]

missile_39 added 2 new tiers that are cheaper, a 500 yen one for getting your name in the credits and a 5,000 yen one for credits + goods. They are almost 30% funded now.

 

[uncreepy]

50% funded, 8 days left. I'm on the edge of my seat

 

[Rylitah]

It's really unfortunate that AHS seems to have distanced themselves from AI JP at least for the time being - a boost from them would have funded the goal by now orz. Real shame since this would be a really unique product if gets funded.

 

[Leon]

Oh man she might not get funded huh... Crossing my fingers for her! It'd be an interesting product for sure.

 

[uncreepy]

61% funded and some absolute madlad bought a foot for $1,000 (probably to try to get the funding to 100% by the deadline)

 

[uncreepy]

Clip of the voice fusion test with Akane (Kansai dialect) from today's live stream:

"Adachi Rei no Kansai-ben ya de" x2
Don't even ask what the 2nd half of the clip says, I listened to this about 7 times and I can only understand half of it. "??? desu. ?? -nai ?? shaberikute(?) desu ne. ??? Adachi Rei desu. ?? INTONEESHON de shabe??"

Apparently, the test is from 2019:


It's really rough... Lots of screeching and is VERY hard to understand her words. Unless this somehow improves by being on the A.I.VOICE engine in the estimated 10ish months to release the bank. Doesn't seem practical for actual Let's Plays and maybe more suitable for joke "I'm a robot" videos or something. Honestly, it seems like only people that can't understand Japanese (because they can't hear how incoherent babbley this sounds), mega TTS Japanese fans (who gotta collect 'em all), and the community of Japanese female robot builders would be hyped by results like this. I tried to search in Japanese for opinions on this demo but could find absolutely nothing.

Sorry if I sound harsh, I am wondering how you all reacted to this clip?


Live stream link:

 

[Rylitah]

I'm willing to give it the benefit of the doubt since it's so old, but that wasn't pleasant to listen to for me either, haha. Even me, talk synth fan as I am... I probably wouldn't get it unless it improved significantly.

There's less than a million yen left to the goal, with two days left. This would have been funded five times over if it were a usual Voiceroid campaign, but looks like it'll be down to the wire here.

 

[___]

Apologies for butting in but this discussion made me remember this one UTAU plug-in that makes talkloids easier and uh....I got curious as to how the results of that plug-in would compare to her wip ai voice vb and I hope it's ok to share the results?

more examples

And and attempt at making her say something longer:


And more fluid vb saying the same thing:


If it's not too undertsandable it apparently is:
至らない点が多いかもしれませんが、頑張りますので、よろしくお 願い　します

Apologies, ik I could have edited those further to make them more natural sounding these are low effort orz but I hope it serves it's purpose, more comptetent examples can be found in the tutorial video:



What is the point? Well... it's always better to use talking vbs for..well....talking because their voice providers recorded with that in mind. Voirecord Itako will always be better for talking than Itako on this plug-in but you can't really make that distinction with Rei...which makes me question whether this project is even worth it....hence I made this comparison to see for myself ...and for you to decide.

 

[uncreepy]

In the live streams by missile_39, he actually does already currently use UTAU to make her speak. It sounds sing-songy, but it actually sounds WAAAY more clear than that A.I.VOICE demo from 2019. I had no clue this plug-in exists, though, so I assume he must be using that like your example

 

[uncreepy]

We can sleep peacefully knowing that Adachi Rei has been funded 104%