iWAX: Interpretable Wav2vec-AASIST-XGBoost Framework for Voice Spoofing Detection

Usage

1. Train Wav2vec 2.0-AASIST Backbone

First, train the Wav2vec 2.0-AASIST model in the aasist directory. This step follows the training process of AASIST GitHub.

Run the following command from the aasist directory:

python main.py --config ./config/wav2vec2_aasist.conf

Make sure the training completes successfully and note the directory where the trained model is saved. This will be used in the next step.

2. Train iWAX

Once the backbone model is trained, you can proceed to train the iWAX classifier:

python iwax.py --save_dir /path/to/save/the/final/model -- w2v2 /path/to/w2v2/you/trained/in/step1

--save_dir: Directory where the final iWAX model will be saved.

--w2v2: Path to the directory or checkpoint of the trained Wav2vec2-AASIST model.

3. Train iWAX with the Sinc Filter

To apply the Sinc filter before classification, use the iwax_sinc.py script:

python iwax_sinc.py --low_freq 128 --high_freq 8000 --time 2 \
    --save_dir /path/to/save/the/final/model \
    --w2v2 /path/to/w2v2/you/trained/in/step1

--low_freq: Lower cutoff frequency (in Hz) for the Sinc filter.

--high_freq: Upper cutoff frequency (in Hz).

--time: n for n/4.