Six Dragons Fly Again: Reviving 15th-Century Korean Court Music with Transformers and Novel Encoding

Danbinaerin Han,Mark Gotham,Dongmin Kim,Hannah Park,Sihun Lee,Dasaem Jeong

2024-08-02

Abstract:We introduce a project that revives a piece of 15th-century Korean court music, Chihwapyeong and Chwipunghyeong, composed upon the poem Songs of the Dragon Flying to Heaven. One of the earliest examples of Jeongganbo, a Korean musical notation system, the remaining version only consists of a rudimentary melody. Our research team, commissioned by the National Gugak (Korean Traditional Music) Center, aimed to transform this old melody into a performable arrangement for a six-part ensemble. Using Jeongganbo data acquired through bespoke optical music recognition, we trained a BERT-like masked language model and an encoder-decoder transformer model. We also propose an encoding scheme that strictly follows the structure of Jeongganbo and denotes note durations as positions. The resulting machine-transformed version of Chihwapyeong and Chwipunghyeong were evaluated by experts and performed by the Court Music Orchestra of National Gugak Center. Our work demonstrates that generative models can successfully be applied to traditional music with limited training data if combined with careful design.

Sound,Artificial Intelligence,Audio and Speech Processing

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: **How to use modern artificial intelligence technology to revive "Jihakpyung" and "Chuwonhaeng" of 15th - century Korean court music and transform them into musical scores that can be played by a six - person orchestra.**

Specifically, the goals of the paper include:

1. **Data Acquisition and Processing**: Obtain Jeongganbo data from existing manuscripts through Optical Music Recognition (OMR) technology and construct a machine - readable data set.
2. **Model Training**: Use a BERT - like masked language model and an encoder - decoder Transformer model to learn and generate melodies in line with the style of traditional Korean court music.
3. **New Encoding Scheme**: Propose a new encoding scheme (Jeonggan - like encoding) to better represent the unique structure of Jeongganbo, especially the position and duration of notes.
4. **Musical Score Generation**: Generate musical scores suitable for six different instruments to play according to a given simple melody, and ensure that the generated musical scores conform to the characteristics of traditional court music in terms of rhythm, melody and grace notes.
5. **Evaluation and Verification**: Verify the quality of the generated musical scores through quantitative indicators and expert subjective evaluation, and finally have the court orchestra of the National Gugak Center of Korea actually play them.

Through these steps, the research team aims to show how to apply deep - learning models to the restoration and innovation of traditional music, especially in the case of limited training data, and combine carefully - designed encoding schemes to achieve high - quality music generation.