🤖 AI Summary
This work addresses the fine-grained detection of joint speech and background audio spoofing by proposing a component-level ensemble method capable of distinguishing among five scenarios: speech-only forgery, background-only forgery, both forged, both genuine, and authentic recordings. The approach integrates four pretrained anti-spoofing models—XLSR-Mamba, DF-Arena, SLS, and TCM-ADD—and incorporates RawBoost data augmentation, multi-head fine-tuning, margin-space score fusion, and a lightweight class-bias calibration strategy. Evaluated in the ICME 2026 ESDD2 Challenge, the method achieves macro F1 scores of 0.7715 and 0.7828 on the evaluation and test sets, respectively, ranking 5th out of 31 participating teams and significantly outperforming the official baseline.
📝 Abstract
This paper describes our submission to the ICME 2026 ESDD2 challenge on environment-aware speech and sound deepfake detection. The task requires five-class classification of audio clips in which speech, environmental sound, both components, or neither component may be spoofed. We propose a component-level ensemble system based on four publicly available pre-trained anti-spoofing models: XLSR-Mamba, DF-Arena, SLS, and TCM-ADD. Each model is fine-tuned on the official CompSpoofV2 development data using three binary heads for original, speech, and environmental sound detection. We further train RawBoost-augmented variants and combine selected checkpoints using margin-space score fusion. A component-wise fusion strategy with lightweight head- and class-bias calibration yields our best configuration, reaching 0.7715 macro-F1 on the evaluation set and 0.7828 macro-F1 on the test set, ranking 5th out of 31 teams in the final ranking phase and substantially outperforming the official baseline.