EBEN: Extreme Bandwidth Extension Network

• This repository is the official implementation of EBEN.
• Visit the project page to listen to audios and visualize some spectrograms.
• Quick start on the project thanks to the Colab demo.

Note: A newer and more performant implementation of EBEN is available in the Vibravox repo.

Requirements

# with Python 3.8.0
pip install pip==24.0
pip install -r requirements.txt

Download French Librispeech

wget https://dl.fbaipublicfiles.com/mls/mls_french.tar.gz
tar xvf mls_french.tar.gz
rm mls_french.tar.gz

Obtain your trained EBEN model

Option 1: use the pre-trained French model discussed in the article

You already have it in the project: generator.ckpt, only 7Mo.

Option 2: train your own model from scratch

python train.py

It will create/refresh generator_retrained.ckpt at the end of each epoch.

Evaluation

python test.py

Results

Our model achieves the following performance on Bandwidth Extension.

In the above Table: format is median (interquartile range). Significantly best values (acceptance=0.05) are in bold. (Note that in the public repo, we only implemented the mean reduction of the metrics)

Cite our work

@ARTICLE{hauret2023configurable_eben_IEEE_TASLP,
  author={Hauret, Julien and Joubaud, Thomas and Zimpfer, V{\'e}ronique and Bavu, {\'E}ric},
  journal={IEEE/ACM Transactions on Audio, Speech, and Language Processing}, 
  title={Configurable EBEN: Extreme Bandwidth Extension Network to Enhance Body-Conducted Speech Capture}, 
  year={2023},
  volume={31},
  number={},
  pages={3499-3512},
  doi={10.1109/TASLP.2023.3313433}}

@inproceedings{hauret2023eben,
  title={EBEN: Extreme bandwidth extension network applied to speech signals captured with noise-resilient body-conduction microphones},
  author={Hauret, Julien and Joubaud, Thomas and Zimpfer, V{\'e}ronique and Bavu, {\'E}ric},
  booktitle={ICASSP 2023 - 2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)}, 
  pages={1--5},
  year={2023},
  organization={IEEE}
  doi={10.1109/ICASSP49357.2023.10096301}}

}