A naturalness-based curriculum and per-sample dynamic temperature reduce speech deepfake detection EER on ASVspoof 2021 DF from 2.45% to 1.88%.
Datasets and metrics In the experiments, all models were trained on the ASVspoof 2019 logical access (LA) dataset [1]
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
eess.AS 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Naturalness-Aware Curriculum Learning with Dynamic Temperature for Speech Deepfake Detection
A naturalness-based curriculum and per-sample dynamic temperature reduce speech deepfake detection EER on ASVspoof 2021 DF from 2.45% to 1.88%.