LISA (Local Implicit representation for Super resolution of Arbitrary scale)

This repository contains the official implementation for LISA introduced in the following paper:

Learning Continuous Representation of Audio for Arbitrary Scale Super Resolution.

The paper can be found at https://arxiv.org/abs/2111.00195.

Environment

• Python 3
• Pytorch 1.6.0
• TensorboardX
• yaml, numpy, tqdm, imageio

If you find out any other dependency is required, please let us know via github issue or pull requests.

Reproducing Experiments

• Read scripts/setup.sh and configs/audio/lisa.yaml carefully for model and experiments setup.

• Prepare datasets to use (such as VCTK).

To train model: run scripts/train_audio.sh

How to cite

If you find our work useful in your research, please cite:

@inproceedings{kim2021lisa,
  title={Learning Continuous Representation of Audio for Arbitrary Scale Super Resolution},
  author={Jaechang Kim and Yunjoo Lee and Seunghoon Hong and Jungseul Ok},
  booktitle={ICASSP},
  year={2022}
}

Reference

This repository is based on previous works below.

• https://github.com/yinboc/liif
• https://github.com/lucidrains/siren-pytorch