Kuroshuu’s Q&A with Yasuo Oda of SSS LLC
This blog post is an English translation of Kuroshuu’s post “東北イタコ/きりたん歌唱DBについて、ずん子陣営に軽く話を聞いた (I casually asked Zunko’s management company about the singing voice databases for Tohoku Itako / Kiritan“ and is translated with permission. Please note that the Q&A has been edited/revised by and is written from Kuroshuu’s perspective.
About Kuroshuu:
A user of close to 30 vocal synthesizer products, who sometimes helps with vocal synth development using their experience. You might recognise Kuroshuu from tweets (such as this one) that went viral during the buzz surrounding AI Kiritan’s release.
The crowdfunding campaign for the development of Tohoku Itako’s singing database has succeeded in meeting its initial goal (and very quickly, at that), and Kuroshuu had the opportunity to talk with Yasuo Oda, president of SSS LLC, the company that manages Tohoku Zunko and her sisters. Here’s their interview.
Regarding Tohoku Itako’s Singing Database
Q : What did you learn from making Tohoku Kiritan’s singing database?
A : Our task was figuring out what should we do with the missing phonemes. But since we have someone who can create phonemes this time around, we should be able to include every phoneme we want. I talked with SHACHI-san to categorize the era of the songs that will be included in the database.
Kuroshuu’s Summary/Thoughts
Tohoku Kiritan’s database contained 50 songs from the voice actress idol group i☆Ris. However that’s not enough to compile every phoneme in the Japanese language. As a result, AI Singer Kiritan sometimes can’t sing lines with minor phonemes such as the “Nya~” sound.
To solve that problem, they invited Meiji University’s Korguchi-san to take part in creating Tohoku Itako’s database. Korguchi-san has the technology to easily create the phonemes that were missing. If it goes well, AI Itako should be able to sing even minor phonemes such as “Nya~”.
Q : What’s your plan going forward?
A : We have begun the tuning process for the songs that will be recorded. Recording will start in August if all goes well. Creating a label for them is a time-consuming task, so it might take 6 months to 1 year overall. We’d like to have the crowdfunding rewards shipped out within 1 month.
Kuroshuu’s Summary/Thoughts
It appears that the process going forward is:
- Song arrangement & schedule adjustment for Kido-san to come to the studio
- Start Recording
- Labeling
- Database release
- Sell the audio material to vocal synthesizer makers
Because the product of this crowdfunding campaign is audio material instead of singing voice synth software, it won’t be a tool that regular users can make sing however they like. SSS are creating the database and are thinking about how they want to make it sing.
Q : Do you have an aim regarding what kind of vocal synth it will be?
A : Whether or not we even have an aim is classified information.
Regarding Tohoku Kiritan’s Singing Database / AI Singer Kiritan
Q : Are there people that use AI singer Kiritan for work purposes?
A : We have received questions from people asking if they can use AI Kiritan for work. We answered that there are no problems with that. We have also heard from people saying how they use her for demo tapes, or for applying to song contests.
Kuroshuu’s Summary/Thoughts
Because they didn’t release the AI Kiritan’s product to the public for commercial purposes, it means that internal usage of her voice bank for private/enterprise use is okay. That’s why we see people using her product for work, or for the demo tape of a contest song.
However, I think it would be really dishonest if someone were to use AI Kiritan for a demo tape, and then lie by saying it’s a live recording when submitting it. I suppose the fact that they would think that they can get away with it is a sign of how good the quality of AI Kiritan’s voice is, which I find to be an interesting phenomenon.
Q : How do you go about asking a voice actress to record for a singing database?
A : We would bring a proposal letter to the voice actress’s agency, explaining what a vocal synthesizer is to the person in charge if they don’t know about it, and then tell them “we want to do this.”
Q : You mean you can explain the concept of an AI singer and then they’ll accept it?
A : Recently there have been many well-known voice actors/actresses that have passed away, and some agencies might find it a shame to lose their voices, like a ‘loss of property.’ Of course there are also people who worry about disruption to their work.
Kuroshuu’s Summary/Thoughts
For a singer, their voice is both their tool of the trade, and also their greatest weapon. I thought that the notion of an AI singer may be frowned upon, but it appears each agency has their own view on it.
To an agency, a voice actor/actress and their voices are important property, and the idea of creating an AI singer for the sake of preserving their voices makes sense. That being said, there may also be the sense of risk from the AI taking their work away.
Q : Are there any copyright concerns regarding the singing database?
A : We make it so that we’re prepared to pay in case copyright holders say something.
Kuroshuu’s Summary/Thoughts
The 50 songs included in Kiritan’s database are all upbeat and lively idol songs. An amendment of the Japanese Copyright Law made it okay to use the songs for research purposes, but they’re still prepared to pay in case there are problems.
Regarding a Doujin Singing Database
Q : What do you think of the recent surge in various people making their own singing database?
A : I think it’s fun. Database creation is tough work because of how draining it is, and not knowing whether it will yield results or not. But the more databases there are, the more research data becomes available, which I think is a good thing. That in and of itself will become Japan’s strength.
Kuroshuu’s Summary/Thoughts
Since the release of Kiritan’s Database, there have been singing database projects coming from UTAU voicebanks’creators. Some of them have even released their completed databases. But at the moment, there is no vocal synth software that can load those databases for the general user to start utilizing them, so if one isn’t careful, there’s the risk of just creating the database, and that’s it.
To some extent, the framework to synthesize the database to the degree of demo voices does exist now, so at the very least, the database can be heard. This is also one of the phenomena that Kiritan’s database created.
Links
Kuroshuu’s Website
Tohoku Itako’s Crowdfunding page
Kuroshuu’s Twitter account