Chord Prediction Model

Introduction

Chord prediction is a common task in music generation and analysis. This project demonstrates the implementation of a chord prediction model using PyTorch. The model uses an Transformer-based architecture to predict chords based on sequences of input notes for real-time harmonizing. The project also contains Max patches for use in AbletonLive for realtime harmonization.

Live Demo

Live Demo

examples

Melody

melody-no-harm.mov

Harmonized Melody

melody-yes-harm.mp4

harmed.mp4

Usage

1. Clone this repository:

   git clone https://github.com/onurio/harmonizing-ai.git
   cd harmonizing-ai

2. Install the required dependencies:

   pip install -r requirements.txt

3. Put all midi files in the midi folder

4. Run the script to create the training data:

   parse-midi.py

5. Run the training notebook:

   HarmonizeTransformer.ipynb

6. Run the prediction script:

   TransformerPredict.py

7. In AbletonLive use the max patch ai-harm.amxd to play midi notes and receive their respective harmonies.

Contributing

Contributions to this project are welcome! You can contribute by:

• Adding new features to the model architecture
• Optimizing the training process
• Enhancing the documentation

Please fork the repository, make your changes, and submit a pull request.

License

This project is licensed under the MIT License.