"A.I. Ch’angga" aims to simulate the Korean music of the early 20th century using machine learning.

'Ch'angga' refers to the music that was educated in the schools in Korea mostly during the Japanese Colonial Era in the early 20th century.

Most songs in the textbook were Japanese songs with translations into Korean. Initially, Korean and Japanese lyrics were written together. But, in the late 1930s, all of the lyrics were changed to Japanese.

‘Ch’angga’ was the mandatory subject that Korean children had to learn. After the liberation of Korea in 1945, the ‘Ch’angga’ education became abolished.

This project has been started from the following question.

“What music would have been created in Korea in the early 20th century if there had not been Japanese colonial era?”

We built a music generation model using the LSTM network, gathering folk songs around the time from China, Russia, Japan, and Korea, and generated music changing the degree of the influence from each country.

For a preliminary test, we built two different models: one trained with all four countries’ music, the other trained with all but Japanese music. The latter one is for simulating the fictional history event where no colonial period existed.

Our preliminary tests suggest that machine learning could be used to simulate the cultural results of an era based on a hypothetical history.

We are planning to expand our database for more accurate prediction and also to make a universal system that has learned all the folk songs around the world letting us explore the composition space in new ways.