GitHub repository storing the source code and resources of this graduate project. Contains the scraper and datasets, trained models, and Jupyter notebooks detailing the project's development together with additional files. This repository is a segment of the larger ChordSeqAI project and primarily serves as a resource for those interested in the technical aspects of AI-driven music composition.
This section includes the dataset and any additional large files used within the notebooks. Uploaded using Git LFS, if you also want to clone this data, use git lfs pull.
Trained PyTorch models, stored using Git LFS.
Jupyter notebooks, the main part of this project. Below is outlined the suggested order in which to go through them.
- Exploratory Data Analysis
- Data Tokenization
- Recurrent Network
- Transformer
- Conditional Transformer
- Classification Transformer
- Style Extraction
- Style Transformer
- Sequence Generation
- Model Evaluation
The LaTeX technical report together with an exported PDF file. To understand this project further, it is recommended to take a look at it.
Source code used throughout the notebooks, parsing the chords and representing sequences of tokens in a readable way. The models directory provides a simple way to apply the trained models elsewhere.
- scraper.py: used to obtain the datasets
- Project Vision.md: written before starting this project, describing its goals
- requirements.txt: a file containing all the Python libraries used in the code, install the dependencies by
pip install -r requirements.txt
The web application can be found in the ChordSeqAI Web App repository.