BTS-E: Audio Deepfake Detection Using Breathing-Talking-Silence Encoder

Code structure:

• biosegment: The Sound Segmentation Model, which uses 3 simple GMM for each class: Brathing, Talking and Silence.

• asvspoof2021/LA/Baseline-Rawnet2-bio: The full pipeline of our work.

Reference

Specially thanks to Hemlata Tak and the ASVspoof organizers for publishing the baseline source code which is easy to follow. We use the Rawnet2 Baseline https://github.com/asvspoof-challenge/2021/tree/main/LA/Baseline-RawNet2 for developing my system.

How to cite

@INPROCEEDINGS{10095927,
  author={Doan, Thien-Phuc and Nguyen-Vu, Long and Jung, Souhwan and Hong, Kihun},
  booktitle={ICASSP 2023 - 2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)}, 
  title={BTS-E: Audio Deepfake Detection Using Breathing-Talking-Silence Encoder}, 
  year={2023},
  volume={},
  number={},
  pages={1-5},
  doi={10.1109/ICASSP49357.2023.10095927}}