Implementation Guidelines

Pre-trained Checkpoints and Supervised Targets

Pre-trained checkpoints and supervised targets (speaker embeddings only) are available at the following link. Since the prosody embedding files are large, we provide the code used to extract the frame-level prosody embeddings.

Link:
https://drive.google.com/drive/folders/1h250-Um5qWo-rpeOE6K_Gdeygy46k3xg?usp=sharing

Available Checkpoints

1. ProSDD trained on ASVspoof 2019
2. ProSDD trained on ASVspoof 2024
3. Baselines trained on ASVspoof 2024: RawNet2; AASIST; XLSR-SLS

These baselines are provided to help the community efficiently use the ASVspoof 2024 dataset.

Evaluation Scores

We also release the evaluation scores for all provided checkpoints.