LVNS-RAVE

This is the codebase for paper LVNS-RAVE: Diversified audio generation with RAVE and Latent Vector Novelty Search accepted as a poster to GECCO 2024.

ACM: https://dl.acm.org/doi/10.1145/3638530.3654432

arxiv: https://arxiv.org/abs/2404.14063

Installation

conda create -n eprior-rave python==3.9
conda activate eprior-rave
pip install -r requirements.txt

Usage

Prepare pretrained RAVEs

• Official RAVE pretrained models
• IIL RAVEs

After download, put the .ts checkpoint file into pretrained_models/.

Modify the LVNS config file

Find a template in eprior/eprior/configs. See eprior/eprior/configs/test.gin for comments of configs.

Start evolution process

cd /eprior/scripts

python run.py --config_path {your_config_file}