The advent of machine generated speech calls for dedicated research to develop countermeasure systems to protect against their misuse through deepfakes. The Speech DF arena leaderboard provides a standardized benchmarking platform for both commercial and open source systems to compare different detection approaches and ranks them using standard metrics. This leaderboard is an evolving inititatie where new systems and attacks can can be added upon request to keep it up to date with the latest advancements in the field. Check out the Submit Your System section to learn how to submit your system.

Below we report the Equal Error Rate (EER %), Accuracy (%) and F1 scores. The table consists of both pooled and average results. Pooled results are computed by using thresholds obtained across all datasets, while average results are computed by simply averaging the dataset level results. We rank the systems according to the Pooled results

If you use Speech DF Arena in your work, it can be cited as:

    @ARTICLE{11345101,
      author={Dowerah, Sandipana and Kulkarni, Atharva and Kulkarni, Ajinkya and Tran, Hoan My and Kalda, Joonas and Fedorchenko, Artem and Fauve, Benoit and Lolive, Damien and Alumäe, Tanel and Magimai.-Doss, Mathew},
      journal={IEEE Open Journal of Signal Processing}, 
      title={Speech DF Arena: A Leaderboard for Speech DeepFake Detection Models}, 
      year={2026},
      volume={},
      number={},
      pages={1-9},
      doi={10.1109/OJSP.2026.3652496}}

Lead Maintainers:
Table: EER (%)
Table: Accuracy (%)
Table: F1 scores