Revolutionary A.I. voice software produces incredible vocals that sound just like a real human

https://ift.tt/2S57gKV Koh Ruide

The day when human singers get replaced by A.I. inches that much closer.

Although virtual singers like Hatsune Miku do a stellar job at singing — even going so far as to hold concert tours around the world — they still retain an unmistakable mechanical tone to their voices.

However, bold research carried out by the International Language Technology Lab in the Nagoya Institute of Technology will change all of that. Working together with speech synthesis company Techno-Speech, the partnership has created a new A.I. technology that is able to reproduce the quality, style, and nuances of human voices more precisely than ever before.

▼ Here is an example of a synthesized voice that is being used today.
Notice that it sounds a little robotic.

▼ And here is the same voice using the new technology.

▼ Pair it with accompanying music and it sounds just like a real singer!

Feeding the A.I. program two hours worth of voice database from a human singer, the team was able to quickly and accurately synthesize vocals by inputting any score with the appropriate lyrics.

This amazing program is not just limited to Japanese, as it is also capable of producing convincing Mandarin and even English voices for songs such as Adele’s Rolling in the Deep and Britney Spears’ Everytime on their official website.

▼ Go to 1:38 here to hear the A.I. version of Adele’s Rolling in the Deep.
Are we sure that’s a computer singing?

これはすごい。
▼AI技術により超高精度な歌声合成を実現：バーチャルシンガーの歌声は人と区別できない時代へhttps://t.co/Z4M6TMoNwL
『株式会社テクノスピーチと名古屋工業大学 国際音声言語技術研究所は、このたび人間の声質・癖・歌い方をこれまでになく高精度に再現できる歌声合成技術を開発』 pic.twitter.com/El2mtAXbyv

— Ken Sugar (@ken_sugar) December 14, 2018

Japanese netizens were blown away by the progress of voice synthesis:

“If you were to say to me that it’s a human singer, I would believe you.”
“Will this mean that singers will lose their jobs? Can we use this to get famous Japanese singers to belt out songs from The Beatles?”
“Sounds great actually!”
“This is revolutionary.”
“It’s fantastic, but it still can’t convey emotions as well as real ones.”

While it is too early to say if this groundbreaking research will spell the end to anyone’s singing career, it does provide a solid foundation for future projects. Once this has progressed sufficiently, however, there may come a day when beautiful A.I.-generated idols with realistic voices become commonplace occurrences in the music industry.

Source: Techno-Speech via Hachima Kiko
Top image: Pakutaso