Xia Yu Yao AI / 夏語遙 AI 聲音庫 Crowdfund Campaign

Xia Yu Yao’s crowdfund has reached approximately 85% funding with 8 days remaining. Keep in mind the numbers are in New Taiwan Dollars (NTD).

The funding goal is ~37k USD, with ~5.7k USD to go. For international backers who want to avoid shipping costs, reward option “F” is the digital voice only and costs ~117USD.

5 Likes
2 Likes

I’m super excited for her AI! If anyone deserves to move on beyond UTAU, Yuyao is definitely one of them

1 Like

It looks like the funding target has been reduced, which means the campaign is less than 1500USD from reaching its goal:

Additionally, a PayPal option has been added (digital voice “option F” only) for people who can’t contribute with a credit card. Gathering additional funds from outside the crowdfund campaign is likely part of what allowed them to reduce the funding target.

3 Likes

Annnnd a big congrats for hitting the goal~! One week to go before it ends!

See you in, hopefully, November 2022!

2 Likes

Is there a sample anywhere of her English ability, I have failed to find anything, either this voicebank or any previous incarnations.

Her existing voice is for UTAU and therefore does not have cross-lingual capabilities.

1 Like

not unless u try hard enough

1 Like

Adding on, Yuyao had Mandarin and Japanese voicebanks specifically. Her Mandarin voicebanks received numerous updates, as well as two appends (INSIDE and OUTSIDE). Yuyao never had a dedicated English voicebank before, so all examples of her in English is her Japanese or Mandarin voicebank manipulated to do so.

2 Likes

Her AI voicebank has yet to be produced or, if it was started at all, hasn’t been shown. You would have to look for her UTAU examples featuring her Mandarin and Japanese banks. Any other language was done using those two voicebanks.

Considering the FAQ, it seems they are planning to make the default language Mandarin, while Japanese and English would be done via Crosslingual. She would not be getting a dedicated Japanese bank this time.

2 Likes

Maybe it’s just me being a bit naive but if they are asking me to fund the project I would think a ‘best effort’ sample could maybe be aired as has been seen on other voices - cross-lingual is still a bit of a dark art I know and some voices seem much better than others :face_with_diagonal_mouth:
OK, the product isn’t finished but they must have some idea that it can/will be done and I would not be suprised if they have a proof of concept in a dark corner somewhere ??
I like what I can find, I have a project I think would suit her range but can I trust her with lead vocals in English? An Xiao (paid v101) has too much accent to sing lead for me, maybe I am just expecting too much from the cross-ling.

Just my two pennys’ worth, that’s all.

I’m not quite sure what you mean by this. Any voice can be used to make an AI voicebank, there is no “proof-of-concept” to be done.

Additionally, due to the nature of AI there’s no way to know exactly how robust her cross-lingual synthesis capabilities will be. If that’s a priority for you then of course you should only purchase native-English voices or finished products that fit your purposes, but it’s not something that can be demonstrated until the product is near completion.

Having a rough demo voice might be neat but also potentially more misleading than helpful, since there’s no guarantee that the final result will align with whatever expectations people might glean from it. We already know her vocal tone and that the native language will be Chinese, that’s enough for me. If it’s not enough for you then you lose nothing by waiting for the completed product.

1 Like

Adding to this, not every company is obligated to show off crosslingual in demonstration. Seika, Ryo, Mo Chen, An Xiao, and Feng Yi did not. Karin did, but they mistakingly used her Japanese when trying to demonstrate Mandarin and didn’t replay it with Mandarin settings and continued on with the stream, so she ended up not having a proper Mandarin demonstration.

The entire point of crosslingual is to make these vocals an option for those languages. Dreamtonics, when introducing the function, said that it is supposed to make the voice fluent in all 3 languages. This benefits everyone who wanted the voice in a specific language but either don’t want to force tune it via phonemes (for tuners), as well as voice providers/companies who were not comfortable producing a voicebank in languages they were not familiar with.

I want to strongly point out that “fluent” does not mean “without accents”. That is something I really disliked when people were putting down “non-native language voicebanks” (Macne Nana English, Meiko English, Luo Tianyi Japanese, etc) and to be honest, when talking about people in real life (especially the latter is rude. People work very hard to learn as much as they can just to communicate in a language they are not native in). It was never promised that the vocals will sing “like a native”, just that they can sing in those languages fluidly and to make it possible for them to do it without having to record those specific languages. I don’t expect Kevin to have “accent-less” Chinese or Japanese. He sounds like an English vocal singing Japanese and Chinese in an American accent. Karin sounds and is a Japanese vocal singing in English and Chinese with a Japanese accent. And in OP’s case, An Xiao sounds like a Chinese vocal singing in English and Japanese with a Chinese accent. And unless they have more data in the other languages aside from default to improve the vocal even further (like Xingchen, and from what was announced, Weina), the accents /might/ be more prominent the less data it has. And to be honest, some people find accents very charming and just want the vocal to be able to sing it in the language they want, accent or not.

Honestly, considering how often I’ve seen comments asking for Utatane Piko English and Gackpoid English/Chinese all those years ago on Vocaloid Wiki and other places, this feature is basically an answer to this kind of request. It even happened for SOLARIA when one of the FAQ was about her getting a Japanese voice database and Eclipsed Sounds said that they couldn’t because Emma Rowley didn’t know the language well enough to produce a quality bank.

For Yuyao, it wouldn’t be unreasonable to expect a Taiwanese accent across all three languages.

Long story short, you buy a vocal built for a specific language and that’s what you get. They’re going to be accented in the other languages. They’re not always going to be “perfect”. If you’re not willing to work with the accent, the vocal probably isn’t for you unfortunately.

5 Likes

Yuyao is finally (re)confirmed to be SynthV AI hahaha

They also opened pre-orders for her digital version if anyone is interested in that.

By the way, here is an example of Mi Yang, the voice actress for Xia Yuyao, speaking as her in a mobile game as a crossover feature. Please keep in mind that this video was posted in March 2015, not very long since she released as an UTAU, so it might not paint an accurate idea of how Yuyao would sound now as a Synthesizer V AI singing vocal. Mi Yang seems to keep herself private, so we don’t really know what her experience is like now compared to back then. She was, however, confirmed to be an aspiring voice actress at the time of Yuyao’s audition back in 2014, so really, who knows? .w./

Unfortunately, I don’t know if there’s any examples of Mi Yang singing as Yuyao as that would be more helpful to figure out how that might sound in Synthesizer V AI (even if it’s an old clip or not).

3 Likes

First recordings have been completed and delivered for data processing. More recordings are expected for this month.

3 Likes

It looks like the second recordings have been completed, and they expect to only need one more session after this.

It’s hard to say what that means as far as progress for the project as a whole, but it is interesting to see that three recording sessions can provide enough samples to create an AI voice.

They also mentioned in today’s tweet that they are planning to make a lite version!

3 Likes

All of the sound samples have been recorded and were sent to Dreamtonics for processing as of last week! Unfortunately, Mi Yang and the recording engineers caught COVID a few days ago, but have recovered now.

Also some progress on the merchandise: the design drafts were sent to the manufacturers and started production.

EDIT:
This was just brought up to me, but it looks like her box was also updated!

2 Likes

Beta testing applications are open until November 25!

Approved beta testers will be expected to provide raw samples (at least 2 minutes long) and feedback about any issues with the voice database. The testing period will run from December 1 to December 10.

I think from this tweet, it can be interpreted that she might not release on time for the expected November period. ^^; They haven’t officially said they’re moving the date, but I think it’s also fair to keep it in mind. Beta testing/demos don’t usually come after the voice releases after all.

4 Likes

I’ve noticed that Dreamtonics has never acknowledged Xia Yu Yao? Like yes, they haven’t done a genbu 2 electric boogaloo, but they said that official business collaboration would be announced on their website/social media, and from what i see (which is nothing), they haven’t?

It’s interesting to see a somewhat strained relationship from the outset…

Yuyao will be delayed a little longer since the Synthesizer V Studio 1.8.0 update went up and introduced Diffusion Probabilistic Models (DPM). As VOICEMITH wants her to be equiped with the newest update to present better singing quality, the beta test version is now being delayed to mid-December, with the final version to be expected in early January 2023.

4 Likes