- Implemented using the MLX framework
- A paired down architecture, inspired by Meta's Encodec model
This project requires Python 3.11.9. To get started, once you've cloned this repository, navigate to the root folower, create a virtual environment and install the requirements:
CONDA_SUBDIR=osx-arm64 conda env create -f environment.yaml
If the command finishes without error, a virtual environment called audio_mlx will be created. Start the virtual environment by running:
conda activate audio_mlx
A dummy dataset consisting of a few audio files is available in the root folder. You can launch a training with:
python train.py
Training loss will be logged in the train_log.log file in the root directory. The default settinsg for training are purely to test the model, to modify for other uses please edit config.yaml .
- Developed and maintained By Charysse Redwood
- Contributions and feature requests are welcomed!