MIDI Chord Recognition via Bar-Level Modeling

This is the codebase for a rule-based MIDI chord recognition system based on dynamic programming. This codebase was used in several papers before, including:

• Wang, Z., Chen, K., Jiang, J., Zhang, Y., Xu, M., Dai, S., ... & Xia, G. (2020). Pop909: A pop-song dataset for music arrangement generation. ISMIR 2020.
• Wang, Z., Wang, D., Zhang, Y., & Xia, G. (2020). Learning interpretable representation for controllable polyphonic music generation. ISMIR 2020.

It can achieve ~80% major/minor chord accuracy on a multi-track pop music MIDI file, like the ones in the RWC Pop dataset.

Installation

pip3 install -r requirements.txt

Usage

python3 main.py ./example.mid 

Example output:

0.000000	2.666640	N
2.666640	5.333280	N
5.333280	7.999920	C#:min
7.999920	10.666560	A:maj9
10.666560	13.333200	A:maj9
13.333200	15.999840	D:maj6(9)
...

Important Notice

1. The algorithm assumes that the beats and downbeats of the input MIDI file (inferred from tempo change events by pretty-midi) are correct. The algorithm uses both beat and downbeat locations to infer the chords.
2. The algorithm might not work well on some music genres, e.g., classical music.

Citation

If you use this repo in your project, please use the following citation (bibtex):

@misc{jiang_midi_2025,
  author       = {Junyan Jiang},
  title        = {{MIDI Chord Recognition via Bar-Level Modeling}},
  year         = {2025},
  howpublished = {\url{https://github.com/music-x-lab/midi-chord-recognition}},
  url          = {https://github.com/music-x-lab/midi-chord-recognition}
}

Misc

To customize the chord dictionary, please refer to chord_class.py.