Search papers, labs, and topics across Lattice.
2
0
4
2
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.
Securing speech systems against deepfakes requires large-scale discriminative SSL models, as shown by the Spoof-SUPERB benchmark, which reveals their superior performance and robustness compared to generative approaches.
