Artificial intelligence recreates voice of the deceased singer


A South Korean TV show used artificial intelligence software to recreate the voice of a singer who died 25 years ago. The program learned details about the artist’s timbre and played a song on stage that was never recorded by him.

The challenge at least curious was carried out by the program “Competition of the Century: AI vs Human”. The musician whose voice was redone was Kim Kwang-seok, who was very successful in the Asian country in the 90s.

Kim sang melancholic songs and died in 1996 after committing suicide. Because of his popularity in South Korea, many of his fans even came up with conspiracy theories that he was murdered. Among the singer’s greatest hits are “Song of My Life” and “In the Wilderness”.

Kim Kwang-seok

The artist was born in Daegu and in the region there are several tributes to him, including a statue

Recreating the voice

The television station SBS was responsible for all the production of the story and to recreate Kim Kwang-seok’s voice she hired the startup Supertone. The software used is called Singing Voice Synthesis (SVS).

About 20 Kim songs were learned by artificial intelligence. “We were very surprised to hear that AI sings better after warming up,” said Kim Min-Ji, producer of the TV show.

In addition to the technique and timbre of the voice, the program also analyzed traces of human emotion in singing and speaking. See, below, the result of recreating the voice of the deceased singer.


According to the producers of the Competition of the Century: AI vs Human, Kim Kwang-seok’s family welcomed the action and considered that the result was as if he had “come back to life”.