Search papers, labs, and topics across Lattice.
This paper investigates supervised contrastive learning (SupCon) for deepfake audio detection using wav2vec2 XLS-R. It explores the impact of cosine vs. angular similarity metrics within the SupCon loss and the effect of negative sample scaling via a cross-batch queue. Results on ASVspoof datasets show that cosine similarity with a delayed queue achieves state-of-the-art performance, while angular similarity performs well even without queued negatives, suggesting it is less reliant on large negative sets.
Angular similarity in supervised contrastive learning can match the performance of cosine similarity for deepfake audio detection, but with significantly less reliance on computationally expensive negative sampling.
Supervised contrastive learning (SupCon) is widely used to shape representations, but has seen limited targeted study for audio deepfake detection. Existing work typically combines contrastive terms with broader pipelines; however, the focus on SupCon itself is missing. In this work, we run a controlled study on wav2vec2 XLS-R (300M) that varies (i) similarity in SupCon (cosine vs angular similarity derived from the hyperspherical angle) and (ii) negative scaling using a warm-started global cross-batch queue. Stage 1 fine-tunes the encoder and projection head with SupCon; Stage 2 freezes them and trains a linear classifier with BCE. Trained on ASVspoof 2019 LA and evaluated on ASV19 eval plus ITW and ASVspoof 2021 DF/LA, Cosine SupCon with a delayed queue achieves the best ITW EER (8.29%) and pooled EER (4.44), while angular similarity performs strongly without queued negatives (ITW 8.70), indicating reduced reliance on large negative sets.
