Audio Dynamic Range Compression (DRC) inversion

Official repository of the paper: Hybrid Model-Driven Inversion of Audio Dynamic Range Compression and Expansion

Authors

• Haoran Sun [email protected]
• Dominique Fourer [email protected]
• Hichem Maaref [email protected]

Affiliations

Laboratoire IBISC (EA 4526), Univ. Evry Paris-Saclay, Évry-Courcouronnes, France

Setup

This repository requires Python 3.10+ and PyTorch 1.10+. Other packages are listed in 'requirements.txt'. To install the requirements in your environment:

pip install -r requirements.txt

Train

To retrain the model, run:

python inference.py -t [task (classification or regression)] -m train -i [input path]

Evaluation

Download the pretrained AST and MEE models, and the test datasets (large): GoogleDrive. To evaluate the pretrained model, run:

python inference.py -t [task (classification or regression)] -m evaluation -i [input path] -o [output path]

Baseline Models

Demucs

HDemucs

De-Limiter

CleanUMamba

Evaluation

To use the 2f-score for evaluation, please see: https://github.com/peladeaucome/PyEvalAudio

License

Distributed under the MIT License.

Acknowledgements

• PyTorch