Other Yamaha to bring back dead singer's voice for NHK tv special using AI

[Exemplar]

I'll leave the translation to much more capable hands

 

[uncreepy]

I'm slightly more than half done with a translation. Will post when ready.

 

[uncreepy]

Note: I translated this in "real time" by writing the English equivalent as I read, so there are some parts that could have sounded more natural. But the article is very long and I didn't want to have to go back and re-read/tweak the sentences or spent a lot of time "properly" translating an article translated as a hobby. Just think of it as either a first draft or that you're reading English how my brain reads Japanese. Plus, I started translating this at 5 am and am very tired.

Article: 当社最新の歌声合成技術『VOCALOID:AI™』で再現 「NHKスペシャル AIでよみがえる美空ひばり（仮）」に技術協力 | VOCALOID ( ボーカロイド・ボカロ ) 公式サイト

Support of the implementation of a new voice library of Hibari Misora
That singer appears as our company's newest singing synthesis technology "VOCALOID: AI™"
Technical cooperation with "NHK Special Hibari Misora resurrected with AI (temporary name)"

Yamaha Corporation has, starting September 29th 9 PM a plan to broadcast the collaboration "NHK Special Hibari Misora resurrected with AI (temporary name)", currently our company is continuing to develop, a singing synthesizer "VOCALOID:AI" using deep learning technology, to recreate the voice of deceased singer Hibari Misora, to realize a live performance of a new song. Furthermore, this will be the first showing to the public of "VOCALOID:AI" and implementing it.

For the effort this time, under the leadership of Japan Broadcasting Corporation (NHK), with the cooperation of many people to implement this, meeting Hibari after 30 years of death, the entertainer Hibari Misora who was top in the song world's new live song actualized through modern-day AI. Reappearing in 4K/3D life-sized hologram the singer reappears, to sing a song produced by Yasushi Akimoto, this is a project to revive Hibari Misora-san's singing voice.

This project at our company was to create the speaking and singing parts during the song the vocalist had in the new song. Specifically, our company used AI technology for our latest singing synthesis technology "VOCALOID:AI", and Hibari Misora'a voice and way of singing, and speaking voice traits faithfully reflected in the singing and speaking parts. The deep learning necessary for the synthesis is, recorded sound sources from her singing and speaking voice while Hibari Misora-san was alive. For the singing voice, there was background music included, but our company used "accompaniment sound removal technology" and extracted the singing voice to create high-quality deep learning data, and actualized high quality synthesis.


VOCALOID:AI Speaking voice + Singing voice > Deep learning > AI that learned Hibari Misora's speaking and singing voice > new song synthesis (song is called "Are Kara (Since Then)")

For the technology we want to use this time, we have been employing a variety of deep learning AI technology which is evolving rapidly, through these latest technologies, we dealt with a challenging effort to resurrect a singer representing the song world of the Showa era.

We think the project this time is, to resurrect the extraordinary vocalist Hibari Misora-san, able to approach the singing voice's essence, our company cultivated musical instruments/audio equipment development/product technology ranging about 130 years. The development of singing synthesis technology has, "a wonderful singing voice that is charming to people and spanning time" which brings forth the possibility of new music, we think this collaboration will express that.

It is planned to broadcast on September 29th the state of this initiative and the new song live.

<Program information>
Program name: "NHK Special Hibari Misora resurrected with AI (temporary name)"
First broadcast: September 29th (Sunday) 9 ~ 9:49 PM
Performance: Yasushi Akimoto, Yoshimi Tendo, Hanae Mori
Program website: ＡＩでよみがえる　美空ひばり（仮）

<About "VOCALOID:AI™" and "VOCALOID™">
"VOCALOID:AI" is our independently developed, singing synthesizer equipped with AI technology. Recording the singer's voice as the objective, the traits like timbre and the way of singing learned though deep learning, that singer's traits and habits and nuance contained in the singing voice, and were able to use them for an arbitrary melody and lyrics.
Our company developed "VOCALOID" singing synthesis in 2003, and are currently spreading close with technology for virtual singers. We advanced dramatically with expressing timbre variation with "VOCALOID:AI" that used AI technology.
Furthermore, due to the official announcement of "VOCALOID:AI", "VOCALOID" will be our company's generic name for singing synthesis technology, "VOCALOID:AI" is therein particularly a thing that uses artificial intelligence.

This page ＡＩでよみがえる　美空ひばり（仮） also got linked to in the article (which shows some behind the scene photos of production work).

Yasushi Akimoto = famous producer for groups such as AKB48, wrote Hibari Misora's last song "Kawa no Nagare no You ni", and the American-born enka singer Jero's debut song "Umi Yuki"
Yoshimi Tendo = enka singer who has been on like... every episode ever of NHK Kouhaku Uta Gassen (Red and White singing contest)


I am wondering if VOCALOID:AI will only be for famous collaborations or if it will be available to the public eventually. It seems kind of lame that people were mourning the death of Vocaloid on Twitter, when they had actually been working on a deep-learning based version of Vocaloid.
Hanae Mori = famous fashion designer who designed for royalty, foreign government officials, operas

 

[uncreepy]

 

[mobius017]

I am wondering if VOCALOID:AI will only be for famous collaborations or if it will be available to the public eventually.

Hard to say. It seems like a lot of work to have done for just a demo of a tech used for other synths someday. Unless they're just going to "resurrect" her without putting the synth into the hands of private producers--if she was really famous, there might be money in that, and it seems a little less strange than putting an actual person's voice in other people's hands and putting that person's actual name on it.

That's a decent post--very sporting of him.

 

[Trevor]

The use of Hibari Misora makes sense. Yamaha has experimented with traditional enka singers before. She seems sorta akin to an Elvis Presley in Japan as well. Yamaha and voctro have been experimenting with "resurrecting" famous singers for private use for a hot minute. Enka would allow the company to showcase the expressiveness of the software while Misora's name will ensure media coverage outside the Vocaloid community. I'm excited to hear how this turns out!

(Side Note: by comparing this to the oddly similar "hint hint" software produced by Voctro, this will most likely be for private use only. Use of the software will probably be expensive and upon request only.)

 

[uncreepy]

Someone uploaded a clip of Hibari Misora singing from NHK:


I think it sounds very realistic and just like her.

 

[peaches2217]

Wait... that was a synthetic voice??? I watched the whole thing and thought it was an actual recording!