In the age of AI, Elvis sings K-pop

Tech

Published: 19 Feb. 2023, 18:10 Updated: 19 Feb. 2023, 19:20

In the age of AI, Elvis sings K-pop

Hosted by Thingsflow and sponsored by Microsoft Korea, Seongsu AI Day conference is held on Thursday in Seongsu, eastern Seoul. [CHOI YEONG-JAE]

In a “Squid Game” (2021) video made for an artificial intelligence (AI) conference, the voice of Gi-hun was dubbed in by an English-speaking voice actor. Using AI-based technology, the voice actor’s voice was modulated into that of the actor portraying the character.

The voice modulation technology was demonstrated at the Seongsu AI Day conference held on Thursday in Seongsu-dong, eastern Seoul.

Hosted by Thingsflow, a subsidiary of Krafton, and sponsored by Microsoft Korea, the event was attended by AI experts from academia and industry, sharing the latest technologies in the field.

Nansy, or Neural Analysis and Synthesis, was one of them.

Nansy is a voice synthesis technology developed by Supertone. Founded in 2020, Supertone is 56 percent owned by HYBE.

Nansy can break down the human voice into separate elements such as tone, pronunciation and accent, and manipulate each factor individually. The software program is faster in processing data than previous voice synthesis programs, and has comparatively clearer pronunciation.

Supertone's voice synthesis AI software, which can synthesize a realistic voice or manipulate the tone of a voice [SUPERTONE]

It can also manipulate vocals in a song.

During the Thursday event, Supertone played an audio sample of Elvis Presley singing a Korean singer’s ballad song. Recreated by Nansy, the legendary rock and roll singer’s voice melted seamlessly into the Korean lyrics of a K-pop song originally sung by Sung Si-kyung.

The global AI market is forecast to quintuple in five years to $407 billion by 2027, according to market tracker MarketsandMarkets. Korea’s AI market will also expand from 2.2 trillion won ($1.67 billion) in 2020 to 27.5 trillion won by 2030.

Amid the accelerated AI race, local tech companies such as Naver and Kakao are speeding up their AI projects. Start-ups are following suit as well, but lack of financing or profitability remains as a major obstacle.

“Though it’s true that venture investments have been shrinking as of late due to the unfavorable market conditions, we are still closely watching the AI sector with a focus on the business’ potential for global expansion,” said Hong Won-ho, CEO of SV Investment.

Koh Kwang-bum, who is leading the gaming and digital native segment at Microsoft Korea, introduced Dall-E, an image generator that can be accessed via Microsoft’s Azure cloud service.

Members of new virtual girl group MAVE: [METAVERSE ENTERTAINMENT]

Developed by OpenAI, Dall-E transforms language prompts into images. For example, if a user types “a puppy landing on the moon” into Dall-E, the program automatically generates an image that meets the user’s command.

“There is a case where a regular user made a children’s book in only two days using ChatGPT and Dall-E,” said Koh.

Microsoft recently introduced Prometheus, a more developed version of ChatGPT language model. Prometheus has been incorporated into the Bing search engine.

Toss Mobile, which entered the mobile virtual network operator (MVNO) market in January, incorporated AI-based technology in its service as well.

“In order to subscribe to the previous MVNO services, consumers had to go through the painful process consisting of nearly 20 stages,” said Ha Dae-woong, chief privacy officer of Viva Republica, operator of internet-only bank Toss.
“We have incorporated automatic speech recognition and other AI-related technologies into our service, and that cut down the subscription process to only two stages.”

Thanks to the speech recognition, consumers do not need to listen to complex and lengthy explanations from the automatic response system (ARS), as the automatic speech recognition system provides a summarized text for the audio explanation.

Neutune, an IT start-up, runs an interactive music editing service named Mix.audio.

“With help of AI technology, users can recreate music by mixing songs from two of their favorite artists,” said Lee Jong-pil, Neutune CEO.

MAVE:, a virtual girl group, made a debut last month. Since first released on Jan. 25, MAVE:’s music video garnered nearly 15 million views on YouTube.

“To build a well-made virtual human, the quality of data used for machine learning is crucial,” said Kang Seong-gu, chief technology officer of Metaverse Entertainment, the agency behind MAVE: and a subsidiary of the game publisher Netmarble.

Ongoing AI research projects were introduced during the Seongsu AI Day as well.

A research team led by Bak Jin-yeong, an assistant professor at the college of computing and informatics at SungKyunKwan University, is focusing on making AI understand human feelings.

The goal is to make AI understand how and why a person is feeling a certain emotion based on the context, taking a step further from merely deducing human feelings from text data.

Hong Sung-eun, professor of information and communication engineering at Inha University, is conducting research on domain adaptation, which involves making a face recognition system identify a person’s facial features not only with a full face but also with their side profiles.

Though AI has seen rapid growth both in terms of market potential and technological development, it still has a long way left to go.

“As it was shown in the experimental case with ChatGPT, AI still only knows what was taught,” said Kim Ju-yeon, professor of computer engineering at the Ulsan National Institute of Science and Technology.

Kim emphasized that “there is still a long way to go.”

“In order for AI to make a judgment based on past experiences like a human brain does, it needs to make constant progress in learning data,” said Kim.

Ha Jung-woo, head of Naver’s AI Lab, also pointed that “data used for AI’s learning process always comes with a risk of copyright issue.s”

The possibility of AI-based software reproducing hateful, discriminatory or inappropriate content it learned during machine learning remains a risk as well.

“Like a real person does, AI needs to listen to people from various backgrounds and cultures, and understand the historical context,” said Oh Hae-yun, professor of computer science at KAIST, addressing the AI bias issue.

“As we continue to strengthen the AI services, I increasingly realize that AI services come with heavy social responsibility,” said Koh of Microsoft.

Koh pointed that “companies need to pay immense attention to preventing unexpected abnormal activities and blocking possible negative influence.”

BY LEE CHANG-KYUN [shin.hanee@joongang.co.kr]