This is the PyTorch implementation of Lite Audio-Visual Speech Enhancement (LAVSE).

We have also put some preprocessed sample data (including enhanced results) in this repository.

The dataset of TMSV (Taiwan Mandarin speech with video) used in LAVSE is released here.

Please cite the following paper if you find the codes useful in your research.

@inproceedings{chuang2020lite,
  title={Lite Audio-Visual Speech Enhancement},
  author={Chuang, Shang-Yi and Tsao, Yu and Lo, Chen-Chou and Wang, Hsin-Min},
  booktitle={Proc. Interspeech 2020}
}

• Ubuntu 18.04
• Python 3.6
• CUDA 10

You can use pip to install Python depedencies.

pip install -r requirements.txt

You can simply enter the command below and the average PESQ and STOI results will show on your terminal pane.

Remember to activate visdom (probably in a screen or tmux) for recording the training loss before bashing the script.

bash run.sh

Go check run.sh if you need further information about the command lines.

The LAVSE work is released under MIT License.

See LICENSE for more details.

• Bio-ASP Lab, CITI, Academia Sinica, Taipei, Taiwan
• SLAM Lab, IIS, Academia Sinica, Taipei, Taiwan