Source code for the paper "A systematic evaluation of adversarial attacks against speech emotion recognition models" [Submitted]
Use conda and the environment file provided
- attacks.ipynb: Run the attacks and the evaluation
- correleations_analysis.ipynb: Perform preliminary analysis on the datasets for identifying most correlated classes
- datasets_preprocessing_phase1.ipynb: Perform pre-processing (step 1)
- datasets_preprocessing_phase2.ipynb: Perform pre-processing (step 2), including data augmentation
- datasets_sr_test.ipynb:
- data_visualization.ipynb:
- defenses-Copy1.ipynb:
- defenses.ipynb:
- hyperparameter_optimization-Copy1.ipynb:
- hyperparameter_optimization.ipynb: Perform the hyper-parameter fine-tuning
- image_specotrograms.ipynb: Create images and audio of original and attacked samples
- metadata_preparation.ipynb:
- model_CNN_LSTM.ipynb: Create the models
- preprocessing_comparison.ipynb:
- random_guess_dummy_classifier.ipynb: Run a random guessing classifier on the data
Here are some audio as examples of the original and attacked samples.
- Male
- Female
- Male
- Female
- Male
- Female
[TODO]