An XLSR-Conformer system trained with Real Emphasis, Fake Dispersion, and multi-class N-pair loss reaches 95.6% in-domain and up to 44.8% out-of-domain source tracing accuracy on MLAAD, versus 83.4% and 26.5% for the Wav2Vec2-AASIST baseline.
Our experiments demon- strated that while the best-performing models (S2, S3) exhibit strong in-domain performance, challenges persist in generaliz- ing to out-of-domain (OOD) data
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.SD 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Unveiling Audio Deepfake Origins: A Deep Metric learning And Conformer Network Approach With Ensemble Fusion
An XLSR-Conformer system trained with Real Emphasis, Fake Dispersion, and multi-class N-pair loss reaches 95.6% in-domain and up to 44.8% out-of-domain source tracing accuracy on MLAAD, versus 83.4% and 26.5% for the Wav2Vec2-AASIST baseline.