From
Send to

AI restores lost music by King Sejong

June 4, 2024 - 16:38 By Hwang Dong-hee
Members of the National Gugak Center perform during the "Reviving 600-year-old Korean Music with Artificial Intelligence" (direct translation) seminar held at the National Gugak Center on Sunday. (National Gugak Center)

The latest technology and ancient music came together at the "Reviving 600-year-old Korean Music with Artificial Intelligence" (direct translation) seminar Sunday, demonstrating how generative artificial intelligence can restore lost ancient music.

According to the records, King Sejong (1397-1450) composed a large work of music, "Bongnaeui," which includes three scores: "Yeomillak," "Chihwapyeong" and "Chwipunghyeong." Among the three, "Yeomillak" has undergone various changes throughout the Joseon era (1392-1910) and continues to be performed today, but the other two pieces exist only in manuscripts, making them impossible to perform.

"The project started with the question, 'What would the lost ancient music sound like if it had been passed down and were still played today?'" said Kim Young-woon, director of the National Gugak Center, which organized the seminar.

The music scores for King Sejong's "Chihwapyeong" (left) and "Chwipunghyeong" (National Institute of Korean History)

"Even if we have the music manuscripts, it's difficult to know the exact rhythm, speed and pitch standards for traditional Korean music, as these were not precisely documented in the traditional music notes. Each note could vary, and the notation for 'sigimsae' (ornamental notes between the main melody notes) is missing, making restoration very challenging once the music is lost," explained Park Jeong-gyeong, a senior researcher at the NGC.

To see how the two lost pieces might be performed if they had been transmitted through time like "Yeomillak," the two teams utilized different methods -- evolutionary algorithms and deep learning -- to restore the music.

Jeong Da-saem, an assistant professor of art and technology at Sogang Univerity, speaks during during the "Reviving 600-year-old Korean Music with Artificial Intelligence" (direct translation) seminar held at the National Gugak Center on Sunday. (National Gugak Center)

First, AI music tech startup CreativeMind utilized evolutionary algorithms -- algorithms using processes that mimic behaviors of living things -- analyzed changes in "Yeomillak" over time and constructed an algorithm to generate the melodies of "Chihwapyeong" and "Chwipunghyeong" as if they had been passed down to the present.

Second, utilizing deep learning, a team from Sogang University led by Jeong Da-saem, an assistant professor of art and technology, employed a dataset of 85 instrument-specific scores published by the NGC between 2014 and 2016. The team recreated "Chihwapyeong" and "Chwipunghyeong" by adhering to the structure of existing traditional Korean music using the learning data.

Both teams said their major challenges were the differences between traditional Korean court music notation and modern staff notation, necessitating a new model for gugak as a whole and sufficient data for the AI to learn from.

"Although the musical scores were created by AI, the various performing techniques demonstrated are performed by musicians based on their interpretation of 'Yeomillak.' While the AI created the scores, it is the musicians who complete the music," said Park.

Director Kim Young-woon of the National Gugak Center speaks at the "Reviving 600-year-old Korean Music with Artificial Intelligence" (direct translation) seminar held at the National Gugak Center on Sunday. (National Gugak Center)

Kim explained the meanings behind the music titles: "Yeomillak" means "to enjoy with the people," "Chihwapyeong" means "to achieve peace," and "Chwipunghyeong" means "to enjoy abundance and prosperity."

"These are the messages King Sejong intended to convey through his music," Kim said.

"The seminar was an opportunity to reflect on King Sejong's spirit. Although there were many challenges due to the unique characteristics of traditional Korean music, we view this as a task to be addressed through future research and experimentation. The meeting of AI and traditional Korean music holds great potential for creating intriguing content in various ways," Kim said.