Pytorch code of the ICASSP 2020 paper "Generating and Protecting Against Adversarial Attacks for Deep Speech-based Emotion Recognition Models", by Zhao Ren, Alice Baird, Jing Han, Zixing Zhang, Björn Schuller.

Database： the Database of Elicited Mood in Speech (DEMoS)

Task: seven-class classification

channels:

• pytorch dependencies:
• matplotlib=2.2.2
• numpy=1.14.5
• h5py=2.8.0
• pytorch=0.4.0
• pip:
  • audioread==2.1.6
  • librosa==0.6.1
  • scikit-learn==0.19.1
  • soundfile==0.10.2

sh runme.sh

In runme.sh, please run the following files for different tasks:

1. feature extraction: utils/features.py

2. training a model, and evaluation: main_pytorch.py

• the folder 'pytorch' is corresponding to vanilla adversarial Training

• the folder 'pytorch-similarity' is corresponding to Similarity-based Adversarial Training

• Please revise the '$BACKEND' to the folder name 'pytorch' or 'pytorch-similarity' in runme.sh, regarding the method which is achieved

If the user referred the code, please cite our paper,

@inproceedings{ren2020generating,

title = {{Generating and protecting against adversarial attacks for deep speech-based emotion recognition models}},

author = {Ren, Zhao and Baird, Alice and Han, Jing and Zhang, Zixing and Schuller, Bj{"o}rn},

address = {Barcelona, Spain},

Booktitle = {Proc.\ ICASSP},

Year = {2020},

pages = {7184--7188}

}

Zhao Ren

Chair of Embedded Intelligence for Health Care and Wellbeing

University of Augsburg

06.07.2020