Implementation of SPEECH FOUNDATION MODEL ENSEMBLES FOR THE CONTROLLED SINGING VOICE DEEPFAKE DETECTION CTRSVDD CHALLENGE 2024, Part of SVDD Challenge

Steps to Implement:

1. Prepare the Dataset:

   • Follow the instructions in the SVDD Challenge CtrSVDD_Utils Repository to manage the dataset.

2. Organize Dataset Structure:

   • Ensure your main directory has the following structure:

     Datasets/
     ├── dev/
     ├── train/
     ├── eval/
     ├── dev.txt
     └── eval.txt
     

3. Set Up the Environment:

   • Create a conda environment using the provided requirements.txt file:

     conda create --name your_env_name --file requirements.txt
     conda activate your_env_name

4. Run Training:

   • Execute the training script by specifying the base directory of the dataset:

     python train.py --base_dir {path_to_Datasets_folder}

   • Additional arguments can be added, such as --algo for the rawboost algorithm:

     python train.py --base_dir {path_to_Datasets_folder} --algo {algorithm_choice}

   • To change the model, modify the model selection directly in the train.py script header.

5. Run Evaluation:

   • Execute the evaluation script by specifying the base directory of the dataset:

     python eval.py --base_dir {path_to_Datasets_folder}

Additional Information:

• Custom Arguments:

  • You can customize various parameters through command-line arguments as needed.
  • Example:

    python train.py --base_dir {path_to_Datasets_folder} --batch_size 64 --epochs 50

• Changing the Model:

  • To use a different model, edit the model import and instantiation in the train.py file.

For further details, refer to the code comments within the scripts.

References

If you would like to use or reference this work in your own research or project, please cite it as follows:

@article{guragain2024speech,
  title={Speech Foundation Model Ensembles for the Controlled Singing Voice Deepfake Detection (CtrSVDD) Challenge 2024},
  author={Guragain, Anmol and Liu, Tianchi and Pan, Zihan and Sailor, Hardik B and Wang, Qiongqiong},
  journal={arXiv preprint arXiv:2409.02302},
  year={2024}
}